Multimodal Large Language Models (MLLMs) employing the Mixture-of-Experts (MoE) structure exhibit encouraging results in visual language tasks. However, they struggle with catastrophic forgetting due to a lack of effective collaboration among experts and negative transfer across tasks. This happens because the router typically employed in MoE for managing expert assignments is inadequate when there are significant shifts in data distribution across various tasks. A drop in the effectiveness of earlier tasks is caused by negative transfer, which occurs due to conflicts in shared knowledge between tasks, disturbing the knowledge already acquired. To address these issues, we propose the Knowledge Space Synergy Framework in Mixture of Experts (KSS-MoE) for Continual Visual Instruction Tuning (CVIT). It dynamically combines the knowledge subspaces of experts to improve the integration of fine-grained complementary knowledge and collaborative abilities of experts, thus addressing the limitations of the basic router. Furthermore, we introduce a general expert that maintains orthogonal subspaces for shared knowledge, enabling effective cross-task knowledge utilization while reducing negative transfer. Extensive experiments conducted on eight CVIT tasks confirm the excellence of KSS-MoE, showcasing its top-tier performance.

Training urban spatio-temporal foundation models that generalize well across diverse regions and cities is critical for deploying urban services in unseen or data-scarce regions. Recent studies have typically focused on fusing cross-domain spatio-temporal data to train unified Transformer-based models. However, these models suffer from quadratic computational complexity and high memory overhead, limiting their scalability and practical deployment. Inspired by the efficiency of Mamba, a state space model with linear time complexity, we explore its potential for efficient urban spatio-temporal prediction. However, directly applying Mamba as a spatio-temporal backbone leads to negative transfer and severe performance degradation. This is primarily due to spatio-temporal heterogeneity and the recursive mechanism of Mamba's hidden state updates, which limit cross-domain generalization. To overcome these challenges, we propose Damba-ST, a novel domain-adaptive Mamba-based model for efficient urban spatio-temporal prediction. Damba-ST retains Mamba's linear complexity advantage while significantly enhancing its adaptability to heterogeneous domains. Specifically, we introduce two core innovations: (1) a domain-adaptive state space model that partitions the latent representation space into a shared subspace for learning cross-domain commonalities and independent, domain-specific subspaces for capturing intra-domain discriminative features; (2) three distinct Domain Adapters, which serve as domain-aware proxies to bridge disparate domain distributions and facilitate the alignment of cross-domain commonalities. Extensive experiments demonstrate the generalization and efficiency of Damba-ST. It achieves state-of-the-art performance on prediction tasks and demonstrates strong zero-shot generalization, enabling seamless deployment in new urban environments without extensive retraining or fine-tuning.

MOTIVATION: Advances in spatial transcriptomics (ST) technologies have made it possible to jointly acquire gene expression and histological image information while preserving spatial coordinates. This breakthrough presents unprecedented opportunities for the precise dissection of spatial heterogeneity in complex tissues. However, existing computational methods remain limited in their capacity for effective integration and synergistic modelling of multimodal ST data. RESULTS: We propose SpatialModal, a multimodal graph learning framework that learns robust joint representations by combining a hierarchical representation strategy with a dual-level contrastive learning mechanism. We perform extensive validation of SpatialModal across diverse ST datasets spanning human and mouse tissues. The results demonstrate that SpatialModal effectively reveals intricate brain architectures in humans and mice, dissects tumour microenvironment heterogeneity in breast cancer, delineates Alzheimer's disease patterns, and characterizes spatiotemporal developmental trajectories within the embryonic heart, underscoring its capability to decipher the spatial heterogeneity of biological tissues. Furthermore, SpatialModal exhibits remarkable versatility and robustness, maintaining superior efficacy even on unimodal datasets devoid of histological images, thereby ensuring its broad applicability across diverse ST platforms. AVAILABILITY AND IMPLEMENTATION: SpatialModal is implemented in Python and is freely available at https://github.com/xingyili/SpatialModal. The source code used in this study has been archived on Zenodo at DOI: https://doi.org/10.5281/zenodo.21264356. All datasets used in this study are publicly available at https://doi.org/10.5281/zenodo.18220735.

In clinical trials, prognostic biomarkers have become essential for guiding treatment decisions after breast cancer surgery. Network-based methods have gained notable attention to reveal marker genes, but many existing methods only focus on a single network, which inevitably neglects the incompleteness of interaction relationships within the network. Even when based upon the multiplex network, most of methods directly integrate the multiplex network into an aggregated network and do not take into account the inherent noise in the biological networks, which can not preserve the topological structure of each original network very well. In this study, we propose a novel method, WEmarker, for breast cancer-specific prognostic analysis. WEmarker reduces the noise level of biological networks and quantifies the probability of interactions between genes, and represents the nodes in the weighted multiplex network as vectors while efficiently retaining the structure information of these networks for identifying prognostic biomarkers. The results show that WEmarker outperforms comparative methods and the case study also demonstrates that biomarkers identified by WEmarker have reliable biological interpretability for breast cancer prognosis.

Abnormal behavior detection is crucial in many fields, such as social networks, financial transactions, and cyber security. However, it poses significant challenges due to the intricate structural evolution of heterogeneous graphs and the need for explainable models. To address these issues, we propose a novel method called Explainable anomalous behavior (edge) detection for dynamic heterogeneous Graphs (ExpGraph). ExpGraph captures relation-aware structural evolution to model temporal behavioral patterns and introduces a prototype alignment mechanism to improve both performance and interpretability. Specifically, prototype alignment enhances detection by en couraging discriminative representations of normal behaviors, which facilitates more accurate identification of anomalies. It also improves interpretability by enabling intuitive explanations through measuring how anomalous behaviors differ from learned normal prototypes. We conduct extensive experiments to evaluate ExpGraph against advanced competitors. It demonstrates that ExpGraph is 16.2% more effective than other methods on average. Moreover, it offers a deeper insight into abnormal behaviors in dynamic heterogeneous graphs. Our code is available at https://github.com/anonymous-123a/ExpGraph.

T cells recognize and eliminate diseased cells by binding their T cell receptors (TCRs) to short endogenous peptides (antigens) presented on the cell surface. Such interactions are central to adaptive immunity, yet current experimental approaches to identify TCR-antigen binding pairs remain labor-intensive and constrained by limited reagents. Here, we propose LoFT- Tcr,a low-rank adaptation (LoRA)-based fine-tuning framework designed for TCR-antigen binding prediction. To capture precise and informative sequence representations, we first fine-tuned the protein large language model ESM-2 on antigen-specific TCR datasets using LoRA. Subsequently, we constructed a heterogeneous interaction graph where nodes encode sequence features and edges indicate TCR-antigen interaction relation-ships. By leveraging a graph learning framework, LoFT- Tcreffectively integrates sequence and topological information to enhance prediction capability. Systematic experiments validated that fine-tuning ESM-2 effectively enhanced the model's capabil-ity to extract discriminative sequence representations, which are critical for accurate TCR specificity prediction. Moreover, LoFT-TCR consistently achieved superior performance compared to state-of-the-art methods on both TCR-antigen binding prediction and TCR specificity discrimination tasks. Experimental results demonstrate that LoFT- Tcrachieves substantial improvements in predictive performance and holds potential for advancing personalized$\mathrm{T}$cell-based immunotherapy. Code of LoFT- Tcris available at https://github.com/sherry-0805/LoFT-TCR

Peptides, as biologically active substances between small molecules and proteins, play a crucial role in drug development due to their high specificity. However, peptides may trigger toxic reactions when exerting a therapeutic effect. Recently, deep learning has emerged as a leading paradigm for peptide toxicity prediction. With the development of protein language models (PLMs), their advantages in protein sequence analysis provide a new technological path for peptide toxicity prediction, but their practical application is limited by the high computational resource requirements. To address the above problems, this paper proposes a peptide toxicity prediction model, PLMTox, that combines PLMs and low-rank adaptation (LoRA) technology. We use protein language models to learn evolutionary information and deep semantic features behind peptide sequences. Simultaneously, PLMTox employs low-rank matrix decomposition to retain the understanding ability of the pre-trained model while realizing the efficient task adaptation of parameters, thus reducing the demand for computational resources. For different datasets, PLMTox (AUC$=0.963$,$\text{SP}=0.960$) exhibits good performance and robustness. These experiment results suggest that PLMTox is expected to promote the safety assessment process of peptide drug development.

Cancer driver genes play an essential role in understanding cancer oncogenesis, tumor progression, and thera-peutic development. The integration of multi-omics data with biological networks has enabled the application of graph deep learning techniques for identifying cancer driver genes. However, most existing methods only use a single biological network as input, inevitably introducing the incompleteness and noise of interactions into models. To address these limitations, we propose MTCDG, a multi-task learning framework for cancer driver gene identification on multi-network and multi-omics data, which can not only enhance the interaction completeness but also enable more comprehensive extraction of graph topological features. The experimental results show the superior predictive performance of MTCDG over other methods. We anticipate that MTCDG will offer new insights for cancer genomic research and can be potentially extended to other areas of biological research in future research. The code of MTCDG is available on github: https://github.com/xingyili/MTCDG.

Similar trajectory retrieval is crucial for maritime trajectory data analysis. However, due to issues such as errors in maritime positioning devices and the accuracy limitations of satellite positioning systems at sea, maritime trajectory data often exhibit characteristics of non-uniform sampling. Existing algorithms struggle to effectively model the irregularity of non-uniformly sampled maritime trajectories, leading to reduced performance in similar trajectory retrieval. In this demonstration, we present STORM, a system designed to effectively retrieve the top-k similar trajectories, which supports both user-specified and automated query settings. STORM utilizes a learnable Fourier-based encoding method to efficiently extract spatiotemporal features from non-uniform trajectories, significantly enhancing the model's performance in similar trajectory retrieval. Our demonstration shows that, compared to state-of-the-art (SOTA) methods, STORM achieves a 41.9% improvement in performance for similar trajectory retrieval on non-uniform maritime data. Our demonstration video is available at https://github.com/itszzzyyy/STORM.

Trajectory data are crucial in intelligent transportation management, road network optimization, and urban mobility analysis. Many downstream applications, such as trajectory prediction and travel time estimation, rely on high-resolution trajectory data. However, real-world trajectories are often sparse due to GPS signal loss and power constraints. Existing trajectory recovery methods often struggle to utilize the latent hierarchical traffic conditions, and they often overlook complex movement semantics. To address these limitations, we propose sparse trajectory recovery with hierarchical dynamic traffic pattern inference (STREAM), a unified framework that collectively infers latent global and local traffic conditions from observed trajectories. By modeling these multi-scale dependencies in its encoder, STREAM enables the decoder to accurately reconstruct missing trajectory points. Additionally, our model effectively captures multi-step movement patterns to enhance the accuracy of next-location inference. Extensive experiments on real-world datasets demonstrate that our model outperforms nine existing competitors with an average improvement of 42.52% in trajectory recovery.

Traditional Chain-of-Thought (CoT) approaches in large language models (LLMs) often miss long-range semantic dependencies. As a result, early reasoning errors may cause subsequent cascading failures. To address these issues, we introduce AACoT, a fine-tuning framework that integrates associative memory and adaptive error correction within the CoT reasoning process. The AACoT memory functions in a dualmode capacity that differentiates entity-level knowledge from relation-level knowledge, facilitating dynamic knowledge interaction and efficient retrieval for reasoning. The adaptive error correction mechanism monitors the reasoning process, backtracks upon error detection, and regenerates the corrected reasoning paths. To improve robustness, a prompt refinement module adjusts short-term memory by collecting frequent error patterns to direct future reasoning, and a memory warm-up strategy loads crucial knowledge in advance of inference to minimize dependency on additional training. In the inference process, AACoT produces several reasoning paths and employs weighted voting to determine the final result. Results from experiments conducted on mathematical reasoning benchmarks reveal significant improvements in accuracy, validating that AACoT provides a clear and effective method for enhancing complex reasoning in foundational LLMs.

Existing research in Stock Trend Prediction (STP) focuses on temporal features extracted from a temporal sequence of stock data with a look-back window, which frequently leads to the omission of important periodic patterns, such as weekly and monthly variations in stock prices. Furthermore, these methods examine stocks individually, ignoring the temporal variation patterns among stocks that share higher-order relationships, like those within the same industry. These relationships typically provide contextual insights into market investments influencing stock price fluctuations. To tackle these issues, we propose a Multi-Scale Temporal Neural Network (MSTNN) framework tailored for STP. This architecture explores the periodic fluctuation behaviors of individual stocks through an innovative 3D convolutional neural network, alongside examining temporal variation patterns of stocks linked to specific industries via a temporal hypergraph attention mechanism. Empirical results from two real-world benchmark datasets show that MSTNN significantly outperforms prior state-of-the-art STP methods. The code of our MSTNN is available at https://github.com/sunlitsong/MSTNN.

Considerable attention has been paid to predicting student performance on exercises. The performance of prior studies is determined by the quality of the trait features of students and exercises. Nevertheless, most of the prior study primarily examines simple pairwise interactions in learning trait features, like those between students and exercises or exercises and concepts, while disregarding the complex higher-order interactions that typically exist among these components, which in turn hinders the prediction results. In this paper, we using an innovative Multi-Channel Graph Contrastive Learning (MCGCL) framework that integrates various high-order interactions for predicting student performance. MCGCL characterizes graph structures reflecting various high-order relationships among students, exercises, and concepts through multiple channels, thereby enhancing the trait features of both students and exercises. Moreover, graph contrastive learning is employed to enhance the representation of trait features acquired from high-order graph structures in diverse views. Extensive experiments on real-world datasets show that MCGCL achieves state-of-the-art results on the task of predicting student performance. The code is available at https://github.com/sunlitsong/MCGCL.

In this paper, we investigate the discovery of temporal anomalous subgraphs in large-scale financial networks, aiming to identify abnormal transaction behaviors among users over time. This task is crucial for the real-time detection of transaction anomalies in financial networks, such as money laundering and trading fraud. However, it poses significant challenges due to the diverse distribution of transactions, the dynamic nature of temporal networks, and the absence of theoretical foundation. To tackle these challenges, we introduce a novel Temporal Anomalous Subgraph Discovery (TempASD) algorithm with theoretical analysis. First, we propose a temporal candidate detection module that quickly pinpoints abnormal candidates by detecting anomalies in both the temporal structure and transaction distribution. Then, we introduce a carefully crafted reinforcement-learning-based refiner to optimize these candidates toward the most abnormal directions. We conducted extensive evaluations against thirteen advanced competitors. TempASD achieves an average improvement of 7x in abnormal degree compared to the state-of-the-art and is efficient in large-scale dynamic financial networks.

Knowledge Graph (KG) reasoning aims to predict missing entities in incomplete triples, which requires adequate structural information to derive accurate embeddings. However, KGs in the real world are not as dense as the idealized benchmarks, where sparse graph structures restrict the comprehensive structural information for superior performance. Although the logical semantics in KGs shows its potential in alleviating the impact of structural sparsity, there still exist some challenges. The deficient supervision and the semantic gap of logic make it difficult to introduce logical semantics in sparse KG reasoning. To this end, we propose a novel KG reasoning approach LoLLM injecting logic with the supervised information supplied by the Large Language Model (LLM), which is proved to be effective in evaluating and scoring. Firstly, LoLLM derives structural embeddings employing a graph convolutional network (GCN) with relation-aware and triple-aware attention. LoLLM secondly constructs reasoning paths instantiated from the first-order logic rules extracted from sparse KGs, and injects the logical semantics by a designed LLM-enhanced tuning strategy. We propose a textual loss (TL) and a logical loss (LL) in the optimization and obtain logical tuning embeddings of KG in this process. Finally, LoLLM fuses structural embeddings from the GCN and logical tuning embeddings from the LLM-enhanced tuning for scoring and incomplete triple prediction. Extensive experiments on two sparse KGs and a benchmark show that LoLLM outperforms state-of-the-art structure-based and Language Model (LM)-augmented baselines. Moreover, the logic rules with corresponding confidences provide explicit explanations as an interpretable paradigm.

Originally proposed as a conceptual metaphor, the Waddington landscape was used to illustrate the directional nature of embryonic development and the relative stability of distinct developmental states. While the Waddington landscape offers a valuable conceptual framework for understanding cellular dynamics, its quantitative reconstruction remains a significant challenge in systems biology and biophysics. Recent methodological advances in single-cell omics technologies, computational modeling approaches, and nonlinear dynamical systems theory have facilitated progress toward quantitative reconstruction of the Waddington landscape, thereby transforming this heuristic metaphor into a predictive theoretical framework. In this review, we summarize the theoretical foundations of the Waddington landscape, categorize current computational and mathematical approaches for the Waddington landscape reconstruction. Meanwhile, we highlight the potential applications and inherent limitations of these approaches in characterizing cellular behaviors, predicting cell fate decisions, and modulating developmental trajectories.

The identification of cancer driver genes is crucial for understanding the complex processes involved in cancer development, progression, and therapeutic strategies. Multi-omics data and biological networks provided by numerous databases enable the application of graph deep learning techniques that incorporate network structures into the deep learning framework. However, most existing methods do not account for the heterophily in the biological networks, which hinders the improvement of model performance. Meanwhile, feature confusion often arises in models based on graph neural networks in such graphs. To address this, we propose a Simplified Graph neural network for identifying Cancer Driver genes in heterophilic networks (SGCD), which comprises primarily two components: a graph convolutional neural network with representation separation and a bimodal feature extractor. The results demonstrate that SGCD not only performs exceptionally well but also exhibits robust discriminative capabilities compared to state-of-the-art methods across all benchmark datasets. Moreover, subsequent interpretability experiments on both the model and biological aspects provide compelling evidence supporting the reliability of SGCD. Additionally, the model can dissect gene modules, revealing clearer connections between driver genes in cancers. We are confident that SGCD holds potential in the field of precision oncology and may be applied to prognosticate biomarkers for a wide range of complex diseases.

Gene regulation varies across different cell types and developmental stages, leading to distinct cellular roles across cellular populations. Investigating cell type-specific gene coexpression is therefore crucial for understanding gene functions and disease pathology. However, reconstructing gene coexpression networks from single-cell transcriptome data is challenging due to artifacts, noise, and data sparsity. Here, we present an efficient method for inference of gene coexpression networks via variance decomposition analysis (GCNVDA) to explore the underlying gene regulatory mechanisms from single-cell transcriptome data. Our model incorporates multiple sources of variability, including a random effect term $G$ to capture gene-level variance and a random effect term $E$ to account for residual errors. We applied GCNVDA to three real-world single-cell datasets, demonstrating that our method outperforms existing state-of-the-art algorithms in both sensitivity and specificity for identifying tissue- or state-specific gene regulations. Furthermore, GCNVDA facilitates the discovery of functional modules that play critical roles in key biological processes such as embryonic development. These findings provide new insights into cell-specific regulatory mechanisms and have the potential to significantly advance research in developmental biology and disease pathology.

The identification of neoantigens is crucial for advancing vaccines, diagnostics, and immunotherapies. Despite this importance, a fundamental question remains: how to model the presentation of neoantigens by major histocompatibility complex class I molecules and the recognition of the peptide-MHC-I (pMHC-I) complex by T cell receptors (TCRs). Accurate prediction of pMHC-I binding and TCR recognition remains a significant computational challenge in immunology due to intricate binding motifs and the long-tail distribution of known binding pairs in public databases. Here, we propose an attention-aware framework comprising TranspMHC for pMHC-I binding prediction and TransTCR for TCR-pMHC-I recognition prediction. Leveraging the attention mechanism, TranspMHC surpasses existing algorithms on independent datasets at both pan-specific and allele-specific levels. For TCR-pMHC-I recognition, TransTCR incorporates transfer learning and a differential learning strategy, demonstrating superior performance and enhanced generalization on independent datasets compared to existing methods. Furthermore, we identify key amino acids associated with binding motifs of peptides and TCRs that facilitate pMHC-I and TCR-pMHC-I binding, indicating the potential interpretability of our proposed framework.

Transposable elements (TEs) are DNA sequences that can move within a genome. They constitute a substantial portion of the eukaryotic genome and play essential roles in gene regulation and genome evolution. Accurate classification of these repetitive elements is crucial for investigating their potential impact on the genome. Over the past few decades, several alignment-based tools have been developed to annotate TE types. While these methods rely heavily on prior knowledge and are often computationally expensive, machine learning-based approaches have been proposed to overcome these limitations. However, most of these approaches fail to capture the multiscale features of TEs, resulting in suboptimal performance. Here, we propose a novel framework called CREATE, which simultaneously integrates the global pattern distribution and the local sequence profile of TEs using Convolutional neural networks and Recurrent neural nEtworks with an Attention mechanism for efficient TE classification. Due to the hierarchical structure of TE groups, we trained nine classifiers corresponding to parent nodes within the class hierarchy. We further applied a top-down hierarchical classification strategy to achieve a more complete classification of unknown TEs. Comprehensive experiments demonstrate that CREATE outperforms existing TE-type annotation methods and achieves superior performance in hierarchical classification tasks. In conclusion, CREATE exhibits great potential for improving the accuracy of TE annotation. The source code and demo data are available at https://github.com/yangqi-cs/CREATE.

Cancer driver genes play a pivotal role in understanding cancer development, progression, and therapeutic discovery. The plenty of accumulation of multi-omics data and biological networks provides a data foundation for graph neural network (GNN) frameworks. However, most existing methods directly concatenate multi-omics data as features, which may lead to limited performance. To address this limitation, we propose deepCDG, a deep graph convolutional network (GCN)-based multi-omics integration model for cancer driver gene identification. The model first employs shared-parameter GCN encoders to extract representations from three omics perspectives, followed by feature integration through an attention layer, and finally utilizes a residual-connected GCN predictor for cancer driver gene identification. Additionally, deepCDG employs GNNExplainer for cancer driver gene module identification. Experimental results demonstrate the effective predictive performance, model robustness, and computational efficiency of deepCDG. Additionally, biological interpretability analysis further validates the reliability of the identification of cancer driver genes of our framework, and the identified gene modules provide profound insights into complex inter-gene relationships and interactions. We believe our method offers enhanced applicability for cancer driver gene identification and could be extended to other biological research fields in future studies.

Cancer remains a significant global health burden, underscoring the need for innovative diagnostic tools to enable early detection and improve patient outcomes. While circulating cell-free DNA (cfDNA) methylation has emerged as a promising biomarker for noninvasive cancer diagnostics, existing methods often face limitations in handling the high-dimensionality of methylation data, small sample sizes, and a lack of biological interpretability. To address these challenges, we propose cfMethylPre, a novel deep transfer learning framework tailored for cancer detection using cfDNA methylation data. cfMethylPre leverages large language model pretrained embeddings from DNA sequence information and integrates them with methylation profiles to enhance feature representation. The deep transfer learning process involves pretraining on bulk DNA methylation data encompassing 2801 samples across 82 cancer types and normal controls, followed by fine-tuning with cfDNA methylation data. This approach ensures robust adaptation to cfDNA's unique characteristics while improving predictive accuracy. Our model achieved superior predictive accuracy compared with state-of-the-art methods, with a weighted Matthews Correlation Coefficient of 0.926 and a weighted F1-score of 0.942. Through model interpretation and biological experimental validation, we identified three novel breast cancer genes-PCDHA10, PRICKLE2, and PRTG-demonstrating their inhibitory effects on cell proliferation and migration in breast cancer cell lines. These findings establish cfMethylPre as a powerful and interpretable tool for cancer diagnostics and biological discovery, paving the way for its application in precision oncology.

Spatial transcriptomics (ST) technology provides gene expression profiles with spatial context, offering critical insights into cellular interactions and tissue architecture. A core task in ST is spatial domain identification, which involves detecting coherent regions with similar spatial expression patterns. However, existing methods often fail to fully exploit spatial information, leading to limited representational capacity and suboptimal clustering accuracy. Here, we introduce MAEST, a novel graph neural network model designed to address these limitations in ST data. MAEST leverages graph masked autoencoders to denoise and refine representations while incorporating graph contrastive learning to prevent feature collapse and enhance model robustness. By integrating one-hop and multi-hop representations, MAEST effectively captures both local and global spatial relationships, improving clustering precision. Extensive experiments across diverse datasets, including the human brain, mouse hippocampus, olfactory bulb, brain, and embryo, demonstrate that MAEST outperforms seven state-of-the-art methods in spatial domain identification. Furthermore, MAEST showcases its ability to integrate multi-slice data, identifying joint domains across horizontal tissue sections with high accuracy. These results highlight MAEST's versatility and effectiveness in unraveling the spatial organization of complex tissues. The source code of MAEST can be obtained at https://github.com/clearlove2333/MAEST.

Breast cancer is a complex disease that arises from multiple factors, including genetics, age, and environmental factors. Prognosis prediction for breast cancer is a challenging task that urgently needs to be addressed. Prognostic biomarkers can aid in predicting clinical outcomes for breast cancer patients, and network-based approaches are frequently employed to identify such biomarkers. However, the accuracy of these approaches based on single source biological network is poor due to incomplete interactions of single biological network. Some network-based approaches that integrate multiple biological networks have not considered network denoising, which may lead to the accuracy of these approaches to be improved. We propose a multi-source fusion identification method named DualMarker for prognostic biomarkers of breast cancer. This method constructs a dual-layer heterogeneous network by integrating multiple biological sources. To decrease the negative effects of incomplete interactions in biological networks, we denoise the constructed network. The ranking of features is obtained by the network propagation algorithm and the initial scoring strategy. Compared with six other network-based methods, DualMarker shows the best performance in six breast cancer datasets. Moreover, we have also demonstrated that the biomarkers identified by DualMarker are of interpretability biologically and closely associated with breast cancer patients' prognosis.

Circular RNAs (circRNAs) play a crucial role in gene regulation and have been implicated in the development of drug resistance in cancer, representing a significant challenge in oncological therapeutics. Despite advancements in computational models predicting RNA-drug interactions, existing frameworks often overlook the complex interplay between circRNAs, drug mechanisms, and disease contexts. This study aims to bridge this gap by introducing a novel computational model, circRDRP, that enhances prediction accuracy by integrating disease-specific contexts into the analysis of circRNA-drug interactions. It employs a hybrid graph neural network that combines features from Graph Attention Networks (GAT) and Graph Convolutional Networks (GCN) in a two-layer structure, with further enhancement through convolutional neural networks. This approach allows for sophisticated feature extraction from integrated networks of circRNAs, drugs, and diseases. Our results demonstrate that the circRDRP model outperforms existing models in predicting drug resistance, showing significant improvements in accuracy, precision, and recall. Specifically, the model shows robust predictive capability in case studies involving major anticancer drugs such as Cisplatin and Methotrexate, indicating its potential utility in precision medicine. In conclusion, circRDRP offers a powerful tool for understanding and predicting drug resistance mediated by circRNAs, with implications for designing more effective cancer therapies.

Road network applications, such as navigation, incident detection, and Point-of-Interest (POI) recommendation, make extensive use of network edge weights (e.g., traveling times). Some of these weights can be missing, especially in a road network where traffic data may not be available for every road. In this article, we study the stochastic weight completion (SWC) problem, which computes the weight distributions of missing road edges. This is difficult, due to the intricate temporal and spatial correlations among neighboring edges. Besides, the road network can be sparse , i.e., there is a lack of traveling information in a large portion of the network. To tackle these challenges, we propose a multi-granularity framework for Region-Wise Graph Completion (RegGC) . To learn coarse spatial correlations among distantly located roads, we construct a region-wise hypergraph neural architecture based on semantic region dependencies. For finer spatial correlations, we incorporate contextual road network properties (e.g., speed limits, lane counts, and road types). Moreover, it incorporates recent and periodic dimensions of road traffic. We evaluate RegGC against 10 existing methods on 3 real road network datasets. They show that RegGC is more effective and efficient than state-of-the-art solutions.

Prerequisite-link Prediction (PLP) aims to discover the condition relations of a specific event or a concerned variable, which is a fundamental problem in a large number of fields, such as educational data mining. Current studies on PLP usually developed graph neural networks (GNNs) to learn the representations of pairs of nodes. However, these models fail to distinguish non-isomorphic graphs and integrate multiscale structures, leading to the insufficient expressive capability of GNNs. To this end, we in this paper proposedk-dimensional Weisferiler-Leman directed GNNs, dubbedk-WediGNNs, to recognize non-isomorphic graphs via the Weisferiler-Leman algorithm. Furthermore, we integrated the multiscale structures of a directed graph intok-WediGNNs, dubbed multiscalek-WediGNNs, from the bidirected views of in-degree and out-degree. With the Siamese network, the proposed models are extended to address the problem of PLP. Besides, the expressive power is then interpreted via theoretical proofs. The experiments were conducted on four publicly available datasets for concept prerequisite relation prediction (CPRP). The results show that the proposed models achieve better performance than the state-of-the-art approaches, where our multiscalek-WediGNN achieves a new benchmark in the task of CPRP.

Predicting protein-ligand binding affinities is a critical problem in drug discovery and design. A majority of existing methods fail to accurately characterize and exploit the geometrically invariant structures of protein-ligand complexes for predicting binding affinities. In this study, we propose Geo-protein-ligand binding affinity (PLA), a geometric equivariant graph representation learning framework with local-global structure awareness, to predict binding affinity by capturing the geometric information of protein-ligand complexes. Specifically, the local structural information of 3-D protein-ligand complexes is extracted by using an equivariant graph neural network (EGNN), which iteratively updates node representations while preserving the equivariance of coordinate transformations. Meanwhile, a graph transformer is utilized to capture long-range interactions among atoms, offering a global view that adaptively focuses on complex regions with a significant impact on binding affinities. Furthermore, the multiscale information from the two channels is integrated to enhance the predictive capability of the model. Extensive experimental studies on two benchmark datasets confirm the superior performance of Geo-PLA. Moreover, the visual interpretation of the learned protein-ligand complexes further indicates that our model offers valuable biological insights for virtual screening and drug repositioning.

Acquiring big-size datasets to raise the performance of deep models has become one of the most critical problems in representation learning (RL) techniques, which is the core potential of the emerging paradigm of federated learning (FL). However, most current FL models concentrate on seeking an identical model for isolated clients and thus fail to make full use of the data specificity between clients. To enhance the classification performance of each client, this study introduces the FDRL, a federated discriminative RL model, by partitioning the data features of each client into a global subspace and a local subspace. More specifically, FDRL learns the global representation for federated communication between those isolated clients, which is to capture common features from all protected datasets via model sharing, and local representations for personalization in each client, which is to preserve specific features of clients via model differentiating. Toward this goal, FDRL in each client trains a shared submodel for federated communication and, meanwhile, a not-shared submodel for locality preservation, in which the two models partition client-feature space by maximizing their differences, followed by a linear model fed with combined features for image classification. The proposed model is implemented with neural networks and optimized in an iterative manner between the server of computing the global model and the clients of learning the local classifiers. Thanks to the powerful capability of local feature preservation, FDRL leads to more discriminative data representations than the compared FL models. Experimental results on public datasets demonstrate that our FDRL benefits from the subspace partition and achieves better performance on federated image classification than the state-of-the-art FL models.

Cancer development is closely linked to the accumulation of mutations in driver genes. Therefore, identification of driver genes is crucial for understanding the molecular basis of cancer. In various types of methods, approches based on Graph Neural Networks (GNNs) framework are one of the most effective tools to identify driver genes, which can also combine biological networks and multi-omics data to further improve the identification accuracy. However, many GNN frameworks often utilize single-order moment to get neighbourhood information for message passing, which ignores the rich distribution information and gene features of neighbouring genes. Besides, when using GNNs model, it is necessary to stack the hidden layers, which can often lead to both over-smoothing and network degradation problems. To overcome these issues, a new framework called DMGNN was proposed for identifying driver genes. To get rich distribution information and gene features, a mix-moment embedding and attention-based feature selection were utilized among neighbouring genes. To solve the problem posed by hidden layer stacking in the GNN model, a deepwalk method was used to learn remote gene effects for the over-smoothing problem, and the resNet-based hidden layer aggregation was employed to mitigate the network degradation issue. Experimental results demonstrate that the proposed model outperforms many existing methods for identifying cancer driver genes, where the AUROC value was achieved at 0.856 in STRING dataset with at least 2 percentage improvement comparing with other models. The DMGNN is freely available via https://github.com/lavendar682/DMGNN.

Breast carcinoma (BRCA) is a leading cause of mortality in women worldwide. Understanding the driving factors behind BRCA initiation, progression, and evolution is crucial. This study proposes a novel method to identify stage-related driving factors in BRCA. By utilizing stage-specific functional interaction networks, the multi-omics features and PPI were integrated. A novel rumor-mongering model is introduced to smooth the stage-specific networks and the gravity model is used to balance gene interactions. The top 100 gravity interactions are identified as driving factors. Through biomolecular and enrichment analyses, these driving factors are shown to play a crucial role in BRCA progression. Furthermore, a hybrid hierarchical evolution network illustrates the stage-evolutionary role of driving factors, while a biological functional evolution network demonstrates functional changes in BRCA progression. The proposed method exhibits superior enrichment performance, particularly within targeted pathways, providing valuable insights into the underlying mechanisms of BRCA.

Graph-level representation learning is important in a wide range of applications. Existing graph-level models are generally built on i.i.d. assumption for both training and testing graphs. However, in an open world, models can encounter out-of-distribution (OOD) testing graphs that are from different distributions unknown during training. A trustworthy model should be able to detect OOD graphs to avoid unreliable predictions, while producing accurate in-distribution (ID) predictions. To achieve this, we present SGOOD, a novel graph-level OOD detection framework. We find that substructure differences commonly exist between ID and OOD graphs, and design SGOOD with a series of techniques to encode task-agnostic substructures for effective OOD detection. Specifically, we build a super graph of substructures for every graph, and develop a two-level graph encoding pipeline that works on both original graphs and super graphs to obtain substructure-enhanced graph representations. We then devise substructure-preserving graph augmentation techniques to further capture more substructure semantics of ID graphs. Extensive experiments against 11 competitors on numerous graph datasets demonstrate the superiority of SGOOD, often surpassing existing methods by a significant margin. The code is available at https://github.com/TommyDzh/SGOOD.

Graph anomaly detection (GAD) has gained increasing attention in recent years due to its critical application in a wide range of domains, such as social networks, financial risk management, and traffic analysis. Existing GAD methods can be categorized into node and edge anomaly detection models based on the type of graph objects being detected. However, these methods typically treat node and edge anomalies as separate tasks, overlooking their associations and frequent co-occurrences in real-world graphs. As a result, they fail to leverage the complementary information provided by node and edge anomalies for mutual detection. Additionally, state-of-the-art GAD methods, such as CoLA and SL-GAD, heavily rely on negative pair sampling in contrastive learning, which incurs high computational costs, hindering their scalability to large graphs. To address these limitations, we propose a novel unified graph anomaly detection framework based on bootstrapped self-supervised learning (named BOURNE). We extract a subgraph (graph view) centered on each target node as node context and transform it into a dual hypergraph (hypergraph view) as edge context. These views are encoded using graph and hypergraph neural networks to capture the representations of nodes, edges, and their associated contexts. By swapping the context embeddings between nodes and edges and measuring the agreement in the embedding space, we enable the mutual detection of node and edge anomalies. Furthermore, BOURNE can eliminate the need for negative sampling, thereby enhancing its efficiency in handling large graphs. Extensive experiments conducted on six benchmark datasets demonstrate the superior effectiveness and efficiency of BOURNE in detecting both node and edge anomalies.

Lung adenocarcinoma, a prevalent subtype of lung cancer, represents one of the most lethal human malignancies. Despite substantial efforts to elucidate its biological underpinnings, the underlying mechanisms governing lung adenocarcinoma remain enigmatic. Modeling and comprehending the dynamics of gene regulatory networks are crucial for unraveling the fundamental mechanisms of lung adenocarcinoma. Conventionally, the cancer is modeled as an equilibrium process based on a time-invariant gene regulatory network to investigate stable cell states. However, the cancer is a nonequilibrium process and the gene regulatory network should be regarded as time-varying in actual. Therefore, a feasible framework was developed to explore the formation and progression of lung adenocarcinoma. On the one hand, to delve into the underlying mechanisms of lung adenocarcinoma formation, the time-invariant gene regulatory network for lung adenocarcinoma was initially undertaken, and the composition of stable cell states was elucidated based on landscape theory. Furthermore, the plasticity of different states was quantified using energy landscape decomposition theory by incorporating cell proliferation. And transition probabilities between different states were defined to elucidate the transition between stable cell states. Additionally, the global sensitivity analysis was performed and a total of three genes and three regulations were identified to be more critical for the formation lung adenocarcinoma, offering a novel strategy for designing network-based therapies for its treatment. On the other hand, the time-invariant gene regulatory network is extended as time-varying to delve into the underlying mechanisms of lung adenocarcinoma progression. The lung adenocarcinoma progression was characterized as four different disease stages based on the mixed states of cell population and the evolutionary direction. And the progressionary mechanism of transition between stages was expounded by evaluating their dynamical transport, with the dynamical transport cost between different stages quantified using Wasserstein metrics.

The evolution of lung adenocarcinoma is accompanied by a multitude of gene mutations and dysfunctions, rendering its phenotypic state and evolutionary direction highly complex. To interpret the evolution of lung adenocarcinoma, various methods have been developed to elucidate the molecular pathogenesis and functional evolution processes. However, most of these methods are constrained by the absence of cancerous temporal information, and the challenges of heterogeneous characteristics. To handle these problems, in this study, a patient quasi-potential landscape method was proposed to estimate the cancerous time of phenotypic states' emergence during the evolutionary process. Subsequently, a total of 39 different oncogenetic paths were identified based on cancerous time and mutations, reflecting the molecular pathogenesis of the evolutionary process of lung adenocarcinoma. To interpret the evolution patterns of lung adenocarcinoma, three oncogenetic graphs were obtained as the common evolutionary patterns by merging the oncogenetic paths. Moreover, patients were evenly re-divided into early, middle, and late evolutionary stages according to cancerous time, and a feasible framework was developed to construct the functional evolution network of lung adenocarcinoma. A total of six significant functional evolution processes were identified from the functional evolution network based on the pathway enrichment analysis, which plays critical roles in understanding the development of lung adenocarcinoma.

Spatial transcriptomics provides valuable insights into gene expression within the native tissue context, effectively merging molecular data with spatial information to uncover intricate cellular relationships and tissue organizations. In this context, deciphering cellular spatial domains becomes essential for revealing complex cellular dynamics and tissue structures. However, current methods encounter challenges in seamlessly integrating gene expression data with spatial information, resulting in less informative representations of spots and suboptimal accuracy in spatial domain identification. We introduce stCluster, a novel method that integrates graph contrastive learning with multi-task learning to refine informative representations for spatial transcriptomic data, consequently improving spatial domain identification. stCluster first leverages graph contrastive learning technology to obtain discriminative representations capable of recognizing spatially coherent patterns. Through jointly optimizing multiple tasks, stCluster further fine-tunes the representations to be able to capture complex relationships between gene expression and spatial organization. Benchmarked against six state-of-the-art methods, the experimental results reveal its proficiency in accurately identifying complex spatial domains across various datasets and platforms, spanning tissue, organ, and embryo levels. Moreover, stCluster can effectively denoise the spatial gene expression patterns and enhance the spatial trajectory inference. The source code of stCluster is freely available at https://github.com/hannshu/stCluster.

Identifying the binding affinity between a drug and its target is essential in drug discovery and repurposing. Numerous computational approaches have been proposed for understanding these interactions. However, most existing methods only utilize either the molecular structure information of drugs and targets or the interaction information of drug-target bipartite networks. They may fail to combine the molecule-scale and network-scale features to obtain high-quality representations. In this study, we propose CSCo-DTA, a novel cross-scale graph contrastive learning approach for drug-target binding affinity prediction. The proposed model combines features learned from the molecular scale and the network scale to capture information from both local and global perspectives. We conducted experiments on two benchmark datasets, and the proposed model outperformed existing state-of-art methods. The ablation experiment demonstrated the significance and efficacy of multi-scale features and cross-scale contrastive learning modules in improving the prediction performance. Moreover, we applied the CSCo-DTA to predict the novel potential targets for Erlotinib and validated the predicted targets with the molecular docking analysis.

SUMMARY: Single-cell multi-omics technologies provide a unique platform for characterizing cell states and reconstructing developmental process by simultaneously quantifying and integrating molecular signatures across various modalities, including genome, transcriptome, epigenome, and other omics layers. However, there is still an urgent unmet need for novel computational tools in this nascent field, which are critical for both effective and efficient interrogation of functionality across different omics modalities. Scbean represents a user-friendly Python library, designed to seamlessly incorporate a diverse array of models for the examination of single-cell data, encompassing both paired and unpaired multi-omics data. The library offers uniform and straightforward interfaces for tasks, such as dimensionality reduction, batch effect elimination, cell label transfer from well-annotated scRNA-seq data to scATAC-seq data, and the identification of spatially variable genes. Moreover, Scbean's models are engineered to harness the computational power of GPU acceleration through Tensorflow, rendering them capable of effortlessly handling datasets comprising millions of cells. AVAILABILITY AND IMPLEMENTATION: Scbean is released on the Python Package Index (PyPI) (https://pypi.org/project/scbean/) and GitHub (https://github.com/jhu99/scbean) under the MIT license. The documentation and example code can be found at https://scbean.readthedocs.io/en/latest/.

BACKGROUND: Structural variations play a significant role in genetic diseases and evolutionary mechanisms. Extensive research has been conducted over the past decade to detect simple structural variations, leading to the development of well-established detection methods. However, recent studies have highlighted the potentially greater impact of complex structural variations on individuals compared to simple structural variations. Despite this, the field still lacks precise detection methods specifically designed for complex structural variations. Therefore, the development of a highly efficient and accurate detection method is of utmost importance. RESULT: In response to this need, we propose a novel method called FindCSV, which leverages deep learning techniques and consensus sequences to enhance the detection of SVs using long-read sequencing data. Compared to current methods, FindCSV performs better in detecting complex and simple structural variations. CONCLUSIONS: FindCSV is a new method to detect complex and simple structural variations with reasonable accuracy in real and simulated data. The source code for the program is available at https://github.com/nwpuzhengyan/FindCSV .

Identifying circRNA-microRNA interactions (CMI) is a significant biomedical issue in recent years. This problem provides insights into using circRNA as biomarkers, developing cancer therapies and producing cancer vaccines. Using computational methods for identification is a more time-efficient and cost-effective approach. In computational methods, using graphs to represent and explore the CMI is a mainstream approach. However, existing relevant methods do not achieve optimal results by utilizing both the semantic information extracted from sequences and the topological information extracted from graph structures. To address this issue, we propose HGLMALLM, a graph contrastive learning method that learns node representation crossing both the semantic domain generated via motif-aware pre-trained LLMs and the topological domain extracted from hierarchical graph structures. Our method effectively addresses the issue in existing Message Passing Neural Network (MPNN) method that edge components losing heterogeneity after multiple iterations. Moreover, this method utilizes the heterogeneity of graph which is extended from the traditional bipartite graph to heterogeneous through the semantic domain . Two commonly used datasets were partitioned based on the distribution of node degrees. Then, we benchmarked our method against existing methods. In the independent testing set evaluation, it achieved a 3 % and 1 % improvement on two datasets. Our method demonstrated the best stability in ten-fold cross-validation on the training set. A test conducted on the peripheral components reveals robust performance of our model. A dataset collected from real scenarios was used to demonstrate the strong predictive ability of our method for identifying unidentified CMI.

Previous Sentiment Analysis (SA) studies have demonstrated that exploring sentiment cues from multiple synchronized modalities can effectively improve the SA results. Unfortunately, until now there is no publicly available dataset for multimodal SA of the stock market. Existing datasets for stock market SA only provide textual stock comments, which usually contain words with ambiguous sentiments or even sarcasm words expressing opposite sentiments of literal meaning. To address this issue, we introduce a Fine-grained Multimodal Sentiment Analysis dataset built upon 1,247 Stock Comment videos, called FMSA-SC. It provides both multimodal sentiment annotations for the videos and unimodal sentiment annotations for the textual, visual, and acoustic modalities of the videos. In addition, FMSA-SC also provides fine-grained annotations that align text at the phrase level with visual and acoustic modalities. Furthermore, we present a new fine-grained multimodal multi-task framework as the baseline for multimodal SA on the FMSA-SC. Data and codes are available athttps://github.com/sunlitsong/FMSA-SC-dataset.git.

This paper proposes a representation learning model for identifying learning disability in resting-state fMRI, with the potential to enhance our understanding of the human brain. The traditional GNN model is used to learn graph representation from fMRI but neglects to consider side information and sparsity. In this paper, we introduce Sparse Spatio-Temporal Graph Neural Networks (SSTGNN) for brain image representation. Specifically, SSTGNN consists of four parts: The ROI Feature Encoder aims to learn temporal ROI features from fMRI, then generate sparse spatial-temporal graphs based on the encoded features, employ GNN for brain image classification, and introduce side information regularization to narrow the gap between the generated graph and prior information. Our model is trained to minimize the cross-entropy (CE) loss. We conducted experiments on the publicly available Autism Brain Imaging Data Exchange dataset. The results demonstrate that the proposed SSTGNN benefits from the introduced side information regularization and sparsity, leading to improved performance in brain classification. This study not only presents an effective fMRI classification model but also has the potential to deepen our understanding of brain intelligence and assist patients with learning disabilities.

Inflammatory bowel disease (IBD) is a complex disease that mainly consists of two subtypes, ulcerative colitis (UC) and Crohn's disease (CD). These two diseases exhibit similar clinical symptoms, leading to a potential misdiagnosis of patients. Accurately assessing the disease status and identifying specific biomarkers are important for the diagnosis and treatment of IBD. Pathways play a significant role in the occurrence and progression of complex diseases, involving the abnormal functionality or regulatory imbalance. Although many methods integrating pathway information have been proposed to evaluate pathway activity, but these methods rarely take into account the topology of pathway networks. Some algorithms based on pathway structure do not de-noise the pathway networks and characterize the disease-specific state of a single sample from the perspective of pathways. In this study, we present a personalized pathway activation inference method (PPA-PS) based on the pathway structure, which utilizes the topology of pathways to evaluate the importance of nodes and quantify the degree of edge disturbance caused by a single disease sample. The results demonstrate that PPA-PS outperforms the compared approaches in terms of classification performance and robustness, indicating its potential as a valuable tool for the pathway biomarker identification and precise diagnosis of IBD.

In the burgeoning field of RNA research, accurate and efficient RNA sequence representation remains a pivotal challenge, exacerbated by the complexity and diversity of RNA sequences. Addressing the critical need for enhanced sequence representation and the issues of sequence context and structural alignment, this study introduces a novel, comprehensive approach. The proposed model seamlessly integrates sequence masking and subsequence consistency optimization, offering a robust solution to the intricate problem of RNA sequence representation. Utilizing the filtered RNAStralign dataset, encompassing 20,923 sequences, the model's performance is rigorously evaluated employing a Support Vector Machine (SVM) for subsequent RNA family classification tasks. Despite the inherent imbalance in RNA family sequence distribution, the model demonstrates exemplary performance, achieving high classification accuracy and AUPRC values across diverse RNA sequence groups. This balanced and unbiased assessment, ensured by the use of AUPRC as an evaluation metric, highlights the model's practical utility for comprehensive RNA sequence analysis and classification. In essence, this research presents a method for enhanced RNA sequence representation and laying a robust foundation for future advancements in the nuanced field of RNA sequence analysis.

In the field of drug discovery, the accurate prediction of drug-target interactions (DTIs) is a critical yet challenging task, hindered by the intricate dynamics of biological systems and molecular interplay. To address this, we propose the DTI-VGAE model, a novel deep learning framework that integrates variational graph autoencoders (VGAE) with a multi-layer perceptron (MLP) for robust DTI prediction. Our approach focuses on three key aspects: learning distinct representations of drugs and proteins from heterogeneous networks, constructing Drug-Protein Pair (DPP) networks to capture the complex interactions, and employing MLP for the final prediction of DTIs. This comprehensive methodology not only enhances the accuracy of DTI predictions but also ensures greater reliability and stability. Validated through extensive 5-fold cross-validation, the DTI-VGAE model consistently outperforms existing methods, achieving superior average AUROC, AUPR scores, and accuracy. The DTI-VGAE model's innovative integration of VGAE and MLP offers a significant advancement in the computational approach to drug discovery, paving the way for more efficient and precise drug development processes.

Identifying the substrates of ubiquitin protein ligase (E3) and deubiquitinases (DUB) contributes to the discovery of potential therapeutic targets for diseases. However, experimental identification of E3/DUB-substrate interactions is costly and time-consuming. Current computational methods for predicting E3/DUB-substrate interactions rely heavily on specific domain knowledge and involve complex and diverse biological data processing. To address this challenge, we proposed a deep learning prediction model, named Deep-USIpred, which predicts E3/DUB-substrate interactions using protein sequences. The proposed Deep-USIpred model encodes protein sequences with a pretrained model and utilizes 1DCNN-BNN deep learning algorithm to make a robust prediction model. We evaluated the performance of the proposed model on real datasets, and our experimental results show that it can achieve excellent prediction performance on the tasks of ESI and DSI. Our proposed method provides a promising alternative for the prediction of E3/DUB-substrate interactions, which has the potential to accelerate drug discovery for various diseases. The source code and dataset are available at https://github.com/PGTSING/Deep-USIpred.

Emerging studies have shown that circular RNAs (circRNAs) are involved in a variety of biological processes and play a key role in disease diagnosing, treating and inferring. Although many methods, including traditional machine learning and deep learning, have been developed to predict associations between circRNAs and diseases, the biological function of circRNAs has not been fully exploited. Some methods have explored disease-related circRNAs based on different views, but how to efficiently use the multi-view data about circRNA is still not well studied. Therefore, we propose a computational model to predict potential circRNA-disease associations based on collaborative learning with circRNA multi-view functional annotations. First, we extract circRNA multi-view functional annotations and build circRNA association networks, respectively, to enable effective network fusion. Then, a collaborative deep learning framework for multi-view information is designed to get circRNA multi-source information features, which can make full use of the internal relationship among circRNA multi-view information. We build a network consisting of circRNAs and diseases by their functional similarity and extract the consistency description information of circRNAs and diseases. Last, we predict potential associations between circRNAs and diseases based on graph auto encoder. Our computational model has better performance in predicting candidate disease-related circRNAs than the existing ones. Furthermore, it shows the high practicability of the method that we use several common diseases as case studies to find some unknown circRNAs related to them. The experiments show that CLCDA can efficiently predict disease-related circRNAs and are helpful for the diagnosis and treatment of human disease.

The emergence of single-cell RNA sequencing (scRNA-seq) technology has revolutionized the identification of cell types and the study of cellular states at a single-cell level. Despite its significant potential, scRNA-seq data analysis is plagued by the issue of missing values. Many existing imputation methods rely on simplistic data distribution assumptions while ignoring the intrinsic gene expression distribution specific to cells. This work presents a novel deep-learning model, named scMultiGAN, for scRNA-seq imputation, which utilizes multiple collaborative generative adversarial networks (GAN). Unlike traditional GAN-based imputation methods that generate missing values based on random noises, scMultiGAN employs a two-stage training process and utilizes multiple GANs to achieve cell-specific imputation. Experimental results show the efficacy of scMultiGAN in imputation accuracy, cell clustering, differential gene expression analysis and trajectory analysis, significantly outperforming existing state-of-the-art techniques. Additionally, scMultiGAN is scalable to large scRNA-seq datasets and consistently performs well across sequencing platforms. The scMultiGAN code is freely available at https://github.com/Galaxy8172/scMultiGAN.

MOTIVATION: Single-cell multimodal assays allow us to simultaneously measure two different molecular features of the same cell, enabling new insights into cellular heterogeneity, cell development and diseases. However, most existing methods suffer from inaccurate dimensionality reduction for the joint-modality data, hindering their discovery of novel or rare cell subpopulations. RESULTS: Here, we present VIMCCA, a computational framework based on variational-assisted multi-view canonical correlation analysis to integrate paired multimodal single-cell data. Our statistical model uses a common latent variable to interpret the common source of variances in two different data modalities. Our approach jointly learns an inference model and two modality-specific non-linear models by leveraging variational inference and deep learning. We perform VIMCCA and compare it with 10 existing state-of-the-art algorithms on four paired multi-modal datasets sequenced by different protocols. Results demonstrate that VIMCCA facilitates integrating various types of joint-modality data, thus leading to more reliable and accurate downstream analysis. VIMCCA improves our ability to identify novel or rare cell subtypes compared to existing widely used methods. Besides, it can also facilitate inferring cell lineage based on joint-modality profiles. AVAILABILITY AND IMPLEMENTATION: The VIMCCA algorithm has been implemented in our toolkit package scbean (≥0.5.0), and its code has been archived at https://github.com/jhu99/scbean under MIT license. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Structural variations (SVs) refer to variations in an organism's chromosome structure that exceed a length of 50 base pairs. They play a significant role in genetic diseases and evolutionary mechanisms. While long-read sequencing technology has led to the development of numerous SV caller methods, their performance results have been suboptimal. Researchers have observed that current SV callers often miss true SVs and generate many false SVs, especially in repetitive regions and areas with multi-allelic SVs. These errors are due to the messy alignments of long-read data, which are affected by their high error rate. Therefore, there is a need for a more accurate SV caller method. RESULT: We propose a new method-SVcnn, a more accurate deep learning-based method for detecting SVs by using long-read sequencing data. We run SVcnn and other SV callers in three real datasets and find that SVcnn improves the F1-score by 2-8% compared with the second-best method when the read depth is greater than 5×. More importantly, SVcnn has better performance for detecting multi-allelic SVs. CONCLUSIONS: SVcnn is an accurate deep learning-based method to detect SVs. The program is available at https://github.com/nwpuzhengyan/SVcnn .

Predicting and understanding student learning performance has been a long-standing task in learning science, which can benefit personalized teaching and learning. This study shows that the progress towards this task can be accelerated by using learning record data to feed a deep learning model that considers the intrinsic course association and the structured features. We proposed a multi-source sparse attention convolutional neural network (MsaCNN) to predict the course grades in a general formulation. MsaCNN adopts multi-scale convolution kernels on student grade records to capture structured features, a global attention strategy to discover the relationship between courses, and multiple input-heads to integrate multi-source features. All achieved features are then poured into a softmax classifier towards an end-to-end supervised deep learning model. Conducting insights into higher education on real-world university datasets, the results show that MsaCNN achieves better performance than traditional methods and delivers an interpretation of student performance by virtue of the resulted course relationships. Inspired by this interpretation, we created an association map for all mentioned courses, followed by evaluating the map with a questionnaire survey. This study provides computer-aided system tools and discovers the course-space map from the educational data, potentially facilitating the personalized learning progress.

Breast cancer is a heterogeneous disease caused by various alterations in the genome or transcriptome. Molecular subtypes of breast cancer have been reported, but useful biomarkers remain to be identified to uncover underlying biological mechanisms and guide clinical decisions. Towards biomarker discovery, several studies focus on genomic alterations that provide differences, while few works concern transcriptomic characterizations that mediate tumor progression. Rather than using differential expression (DE) or weighted network analysis, we propose a feature selection method, dubbed GLassonet, to identify discriminative biomarkers from transcriptome-wide expression profiles by embedding the relationship graph of high-dimensional expressions into the Lassonet model. GLassonet comprises a nonlinear neural network for identifying cancer subtypes, a skipping fully connected layer for canceling the connections of hidden layers from input features to output categories, and a graph enhancement for preserving the discriminative graph into the selected subspace. First, an iterative optimization algorithm learns model parameters on the TCGA breast cancer dataset to investigate the classification performance. Then, we probe the distribution patterns of GLassonet-selected gene sets across the cancer subtypes and compare them to gene sets outputted from the state-of-the-art. More profoundly, we conduct the overall survival analysis on three GLassonet-selected new marker genes, i.e., SOX10, TPX2, and TUBA1C, to investigate their expression changes and assess their prognostic impacts. Finally, we perform the enrichment analysis to discover the functional associations of the GLassonet-selected genes with GO terms and KEGG pathways. Experimental results show that GLassonet has a powerful ability to select the discriminative genes, which improve cancer subtype classification performance and provide potential biomarkers for cancer personalized therapy.

Data warehouse and OLAP (Online Analytical Processing) are effective tools for decision support on traditional relational data and static multidimensional network data. However, many real-world multidimensional networks are often modeled as temporal multidimensional networks, where the edges in the network are associated with temporal information. Such temporal multidimensional networks typically cannot be handled by traditional data warehouse and OLAP techniques. To fill this gap, we propose a novel data warehouse model, named$\mathsf {Temporal{ }\; Graph{ }\; Cube}$, to support OLAP queries on temporal multidimensional networks. Through supporting OLAP queries in any time range, users can obtain summarized information of the network in the time range of interest, which cannot be derived by using traditional static graph OLAP techniques. We propose a segment-tree based indexing technique to speed up the OLAP queries, and also develop an index-updating technique to maintain the index when the temporal multidimensional network evolves over time. In addition, we also propose a novel concept called$\mathsf {similarity{ }\; of{ }\; snapshots}$which shows a strong correlation with the efficiency of indexing technique and can provide a good reference on the necessity of building the index. The results of extensive experiments on two large real-world datasets demonstrate the effectiveness and efficiency of the proposed method.

Cross-Modal Zero-Shot Hashing (CMZSH) is an important image retrieval technique, e.g., Text Based Image Retrieval. Most of existing CMZSH methods mainly use semantic attributes as guidance to generate hash codes for both the images and texts of seen and unseen categories. However, existing CMZSH methods only focus on learning global attribute vectors and hash codes for images, which mixes up information of complex semantics and background clutters, and thus impedes the retrieval performance. To solve this issue, we propose an Attribute-Guided Multiple Instance Hashing (AG-MIH) network for CMZSH, where each instance represents one image region. Instead of generating global image hash codes, AG-MIH can effectively learn instance-level hash codes based on instance attributes. To improve the attribute learning for instances, AG-MIH can exploia novel 2-D Category-Attribute Relation (CAR) layer, which uses different matching templates to model the relationships between each instance and the attributes for different categories. Under the guidance of semantic attributes, AG-MIH can effectively learn hash codes for each visual instance and texts by a Multi-stream Instance Hashing Refinement (MIHR) procedure. In the MIHR, the pseudo supervisions for the instance-level attributes and hash codes in each stream are from its proceeding stream. Empirical studies on benchmark datasets show that AG-MIH achieves state-of-the-art performance on both cross-modal and single-modal zero-shot image retrieval tasks.

Current Heterogeneous Network Embedding (HNE) models can be roughly divided into two types, i.e., relation-aware and metapath-aware models. However, they either fail to represent the non-pairwise relations in heterogeneous graph, or only capable of capturing local information around target node. In this paper, we propose a metapath based multilevel graph attention networks (MMAN) to jointly learn node embeddings on two substructures, i.e., metapath based graphs and hypergraphs extracted from original heterogeneous graph. Extensive experiments on three benchmark datasets for node classification and node clustering demonstrate the superiority of MMAN over the state-of-the-art works.

Long noncoding RNAs (lncRNAs) can regulate the expression of protein-coding genes (PCGs) to cause disease. Identifying lncRNA-PCG associations (LGAs) is beneficial in revealing the pathogenic mechanism of lncRNA. Nevertheless, it remains challenging due to the heterogeneity of lncRNA expression and the complexity of its regulatory patterns. Biological experiments have been designed to identify LGAs, but they cannot be used on a large scale due to time and financial constraints. Therefore, the design of computational methods becomes crucial for LGA research. Here, we propose a new computational model, DNNMC, to reveal potential LGAs based on deep neural networks and inductive matrix completion using association and multi-omics data. We first integrated LGA and multi-omics similarity to construct lncRNA and PCG similarity networks. Subsequently, deep graph convolutional networks were used for feature learning of lncRNAs and PCGs. These learned features and the known LGA matrix were finally used as input to the inductive matrix completion module for predicting potential LGAs. Experimental results on three datasets demonstrated that DNNMC outperformed other machine learning methods in predicting LGA relationships. Furthermore, multi-omics features were shown to improve the performance of LGA identification. In conclusion, we propose a new LGA prediction method, DNNMC, which can effectively complete the LGA prediction task and help to reveal the regulatory mechanism of lncRNAs in diseases.

The prognosis of breast cancer is challenging, which is an urgent problem to be solved. The prognostic biomarkers for breast cancer can help us predict the clinical outcomes of patients, and network-based methods are widely introduced to find prognostic biomarkers. According to the difference of input biological data, existing network-based biomarker prediction methods are mainly classified into two types: integrating single-source network or multi-source networks. However, the interactome of single-source network remains incomplete, and biological networks are noisy, which will hamper the network-based identification accuracy of biomarkers. In this study, we introduce a multi-source fusion method, DualMarker, which integrates multiple biological information sources and constructs a dual-layer heterogeneous network by fast network embedding. Next, we introduce a network enhancement method to denoise the constructed dual-layer heterogeneous network, and we implement network propagation algorithm on the constructed dual-layer heterogeneous network to rank the features. After comparing with competitive methods, we find that DualMarker substantially outperforms these methods. In addition, we verify that the biomarkers identified by DualMarker are closely related to the prognosis of breast cancer patients.

Circular RNAs (circRNAs) have important effects on various biological processes, and their dysfunction is closely related to the emergence and development of diseases. Identifying the associations between circRNAs and diseases is helpful in analyzing the pathogenesis of diseases. Therefore, it is necessary to develop effective computational methods for predicting circRNA-disease associations. Here, we present a computational model called HRCDA to predict associations between circRNA and disease based on heterogeneous graph representation. Firstly, an integrated network of circRNA functional similarity is built by Random Walk with Restart in the view of biological functions of circRNA. Then, a heterogeneous graph of circRNAs and diseases is constructed with known circRNA-disease associations. Finally, we design a heterogeneous graph representation learn model based on Graph Auto-Encoder (GAE) to predict circRNA-disease associations. Experiments have shown that the proposed method perform better than existing state-of-the-art methods and can be an effective tool to predict potential disease-related circRNAs.

Disease gene identification has provided valuable insights into illuminating the molecular mechanisms underlying complex diseases. And it has been shown that novel drugs with genetically supported targets were more likely to be successful in clinical trials. In recent years, multiple graph machine learning-based methods for this purpose have been proposed. However, those methods were mainly based on various well-established biological molecular networks, while seldomly considering the curated biological annotations of genes. To fill this gap, we aim to integrate the gene ontology annotations (GOA), including the biological process (BP), the cellular component (CC), and the molecular function (MF), into the process of disease gene prediction. Our method treated the GOA as a hypergraph and used the hypergraph-based embedding technique to extract the deep features underlying gene annotations. Besides, we also extracted gene features from the protein-protein interaction (PPI) network using graph representation learning methods. The convolutional neural network (CNN) framework was followed to fuse the features extracted from the two networks and make the final prediction. Experiments on a range of diseases have demonstrated the accuracy and robustness of our method. The average area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC) reached 0.85 and 0.79, respectively. Besides, the hypergraph-based gene ontology embedding can be generalized to other bioinformatics applications.

The expression quantitative trait loci (eQTL) analysis has become important for understanding the regulatory function of genomic variants on gene expression in a tissuespecific manner and has been widely applied across species from microbes to mammals. Current eQTL studies mainly focus on the simple one-to-one regulation between variant and gene. Recent research have demonstrated there are also more complex regulatory patterns between eQTLs and genes. However, there is a lack of studies and relevant methods to systematically discover the regulatory patterns between multiple eQTLs and multiple genes. In this regard, this study has proposed a novel computational framework, called eQTLMotif, to discover regulation patterns of eQTLs in a many-to-many manner. This framework mainly consists of two steps: (1) construct a novel eQTL regulatory network by integrating bipartite eQTL network, eQTL mediation effects, and gene regulatory network; (2) perform motif mining through exactly enumerating frequently appeared eQTL regulatory structures. Based on this framework, we for the first time systematically investigated the eQTL regulatory patterns in the human frontal cortex based on a large cohort of postmortem human brains. Experiments have demonstrated that our framework can effectively reveal novel eQTL regulatory patterns. And some are in similar structure to the existing gene regulation patterns, such as feed-forward loop (FFL)-like motif, single input module (SIM)-like motif, and dense overlapping regulons (DOR)- like motif. Our method and findings will further enhance the understanding of regulatory mechanisms of eQTLs in multiple tissues and species.

This paper proposes a representation learning model to identify task-state fMRIs for knowledge-concept recognition, which has the potential to model the human cognitive expression system. The traditional CNN-LSTM is usually employed to learn deep features from fMRIs, where CNN aims at extracting the spatial structure and LSTM accounts for the temporal structure. However, the manifold smoothness of the latent features caused by the fMRI sequence is often ignored, leading to unsteady data representation. In this paper, we model latent features as a hidden Markov chain and introduce a Markov-guided Spatio-Temporal Network (MSTNet) for brain image representation. Concretely, MSTNet has three parts: CNN that aims to learn latent features from 3D fMRI frames where a Markov Regularization enforces the neighborhood frames to have similar features, LSTM integrates all frames of an fMRI sequence into a feature vector and fully connected network (FCN) that is to implement the brain image classification. Our model is trained towards minimizing the cross entropy (CE) loss. Our experiment is conducted on the brain fMRI datasets achieved by scanning college students when they were learning five concepts of computer science. The results show that the proposed MSTNet can benefit from the introduced Markov regularization and thus result in improved performance on the brain activity classification. This study not only shows an effective fMRI classification model with Markov regularization but also provides the potential to understand brain intelligence and help patients with language disabilities.

Federated learning (FL) allows a large number of clients to improve their respective models through training a shared global model. However, passing the same global model is not conducive to the training of a few clients and leads to a large loss of localization information. In practical, there are often some prior information that can be shared between clients. Our study takes into account the use of such prior information to calculate a personalized global model for each client, resulting in an enhanced personalized federated learning framework, dubbed PerFL for short, that takes advantage of available client features that can be shared with other clients. More specifically, PerFL calculates the incidence matrix of all involved clients by using the permitted shareable side information and then updates the local models by using their similar clients instead of all clients. Employing the neural network as the classification model, PerFL learns the parameter matrices at each client in an iterative manner. On three publicly available image datasets, PerFL can benefit from the employed similarity and achieve an improved classification performance in comparison with the state-of-the-art FL models.

This paper studies the problem of developing contrastive learning into the privacy-protected federated learning (FL), which is to achieve more data samples for model training. The existing methods usually encourage the global model and local models in FL to be the same one, often ignoring the data heterogeneity of the clients. In this paper, we proposed a method of personalized federated contrastive learning to improve the FL model performance for each client’s task, by learning a global representation and a local representation simultaneously. Our method is a novel FL framework that borrows the scheme of contrastive learning (CL), where one CL branch is the global model while the other branch is the local model divided into a share part and a personalized part. The proposed model is then trained by maximizing the agreement between the global model and the sharing part of the local model and meanwhile minimizing the agreement between the global model and the personalized part. We conducted evaluations on three public datasets for federated image classification. The results show that the proposed method can benefit from the personalization of local models and thus achieve better accuracy in comparison with the state-of-the-art FL models.

Fake news detection is a challenging problem due to its tremendous real-world political and social impacts. Recent fake news detection works focus on learning news features from News Propagation Graph (NPG). However, little attention is paid to the issues of both authenticity of the relationships and topology imbalance in the structure of NPG, which trick existing methods and thus lead to incorrect prediction results. To tackle these issues, in this paper, we propose a novel Topology imbalance and Relation inauthenticity aware Hierarchical Graph Attention Networks (TR-HGAN) to identify fake news on social media. Specifically, we design a new topology imbalance smoothing strategy to measure the topology weight of each node. Besides, we adopt a hierarchical-level attention mechanism for graph convolutional learning, which can adaptively identify the authenticity of relationships by assigning appropriate weights to each of them. Experiments on real-world datasets demonstrate that TR-HGAN significantly outperforms state-of-the-art methods.

The joint problem of imbalance samples and noisy labels challenges the current text classifiers in real-world applications. Existing approaches are mostly devoted to handling either former or latter while fail to manage the fused issue. This paper introduces a novel weakly-supervised framework, dubbed WeSt-coin, to take into account the sensitivity cost on misclassifications between classes and seek seed words towards noisy-label corrections. After BERT that creates a contextualized corpus, WeStcoin learns a predicted label vector from the contextualized samples and meanwhile calculates a pseudo probability vector from seed words, and then projects the concatenated representation into an output space, followed by multiplying by a cost-sensitive matrix. WeStcoin is ultimately trained to decrease the residual between the model outputs and the noisy labels, where seed words are also updated in an iterative manner. Extensive experiments and ablation studies on two public text datasets demonstrate that the proposed model outperforms the state-of-the-art model in the text classification with imbalance samples and noisy labels. Codes are made available at https://github.com/ypzhaang.

Recommendations in the educational environment aim to help learner access their personalized demands efficiently. Unlike commodity recommendation, limited to the ethics of pedagogy and the high cost of bad recommendations, the credibility and interpretability of the education recommendation system are more worthy of attention to achieve recommendation accuracy. However, few studies focused on the interpretability of recommendations. Thus, this study proposes an Open Recommendation framework for Interpretability based on the Bayesian principal component analysis (PPCA), ORec4Int. ORec4Int helps learners understand the recommendation by building a mapping between educational resources and the latent factors/features of learners. The interpretability will enhance his/her trust in the education recommendation system. Finally, We not only evaluate the recommendation performance of ORec4Int based on one real-world dataset but also compared its performance in interpretability and the education expert solution. Results show that ORec4Int can approach the performance of education expert solutions. Ultimately, ORec4Int is faster, more efficient, and less costly.

Long non-coding RNAs (lncRNAs) can disrupt the biological functions of protein-coding genes (PCGs) to cause cancer. However, the relationship between lncRNAs and PCGs remains unclear and difficult to predict. Machine learning has achieved a satisfactory performance in association prediction, but to our knowledge, it is currently less used in lncRNA-PCG association prediction. Therefore, we introduce GAE-LGA, a powerful deep learning model with graph autoencoders as components, to recognize potential lncRNA-PCG associations. GAE-LGA jointly explored lncRNA-PCG learning and cross-omics correlation learning for effective lncRNA-PCG association identification. The functional similarity and multi-omics similarity of lncRNAs and PCGs were accumulated and encoded by graph autoencoders to extract feature representations of lncRNAs and PCGs, which were subsequently used for decoding to obtain candidate lncRNA-PCG pairs. Comprehensive evaluation demonstrated that GAE-LGA can successfully capture lncRNA-PCG associations with strong robustness and outperformed other machine learning-based identification methods. Furthermore, multi-omics features were shown to improve the performance of lncRNA-PCG association identification. In conclusion, GAE-LGA can act as an efficient application for lncRNA-PCG association prediction with the following advantages: It fuses multi-omics information into the similarity network, making the feature representation more accurate; it can predict lncRNA-PCG associations for new lncRNAs and identify potential lncRNA-PCG associations with high accuracy.

Single-cell technologies provide us new ways to profile transcriptomic landscape, chromatin accessibility, spatial expression patterns in heterogeneous tissues at the resolution of single cell. With enormous generated single-cell datasets, a key analytic challenge is to integrate these datasets to gain biological insights into cellular compositions. Here, we developed a domain-adversarial and variational approximation, DAVAE, which can integrate multiple single-cell datasets across samples, technologies and modalities with a single strategy. Besides, DAVAE can also integrate paired data of ATAC profile and transcriptome profile that are simultaneously measured from a same cell. With a mini-batch stochastic gradient descent strategy, it is scalable for large-scale data and can be accelerated by GPUs. Results on seven real data integration applications demonstrated the effectiveness and scalability of DAVAE in batch-effect removing, transfer learning and cell-type predictions for multiple single-cell datasets across samples, technologies and modalities. Availability: DAVAE has been implemented in a toolkit package "scbean" in the pypi repository, and the source code can be also freely accessible at https://github.com/jhu99/scbean. All our data and source code for reproducing the results of this paper can be accessible at https://github.com/jhu99/davae_paper.

Brain disease gene identification is critical for revealing the biological mechanism and developing drugs for brain diseases. To enhance the identification of brain disease genes, similarity-based computational methods, especially network-based methods, have been adopted for narrowing down the searching space. However, these network-based methods only use molecular networks, ignoring brain connectome data, which have been widely used in many brain-related studies. In our study, we propose a novel framework, named brainMI, for integrating brain connectome data and molecular-based gene association networks to predict brain disease genes. For the consistent representation of molecular-based network data and brain connectome data, brainMI first constructs a novel gene network, called brain functional connectivity (BFC)-based gene network, based on resting-state functional magnetic resonance imaging data and brain region-specific gene expression data. Then, a multiple network integration method is proposed to learn low-dimensional features of genes by integrating the BFC-based gene network and existing protein-protein interaction networks. Finally, these features are utilized to predict brain disease genes based on a support vector machine-based model. We evaluate brainMI on four brain diseases, including Alzheimer's disease, Parkinson's disease, major depressive disorder and autism. brainMI achieves of 0.761, 0.729, 0.728 and 0.744 using the BFC-based gene network alone and enhances the molecular network-based performance by 6.3% on average. In addition, the results show that brainMI achieves higher performance in predicting brain disease genes compared to the existing three state-of-the-art methods.

Quantitative trait locus (QTL) analyses of multiomic molecular traits, such as gene transcription (eQTL), DNA methylation (mQTL) and histone modification (haQTL), have been widely used to infer the functional effects of genome variants. However, the QTL discovery is largely restricted by the limited study sample size, which demands higher threshold of minor allele frequency and then causes heavy missing molecular trait-variant associations. This happens prominently in single-cell level molecular QTL studies because of sample availability and cost. It is urgent to propose a method to solve this problem in order to enhance discoveries of current molecular QTL studies with small sample size. In this study, we presented an efficient computational framework called xQTLImp to impute missing molecular QTL associations. In the local-region imputation, xQTLImp uses multivariate Gaussian model to impute the missing associations by leveraging known association statistics of variants and the linkage disequilibrium (LD) around. In the genome-wide imputation, novel procedures are implemented to improve efficiency, including dynamically constructing a reused LD buffer, adopting multiple heuristic strategies and parallel computing. Experiments on various multiomic bulk and single-cell sequencing-based QTL datasets have demonstrated high imputation accuracy and novel QTL discovery ability of xQTLImp. Finally, a C++ software package is freely available at https://github.com/stormlovetao/QTLIMP.

In the originally published version of this manuscript, there was an error in the Funding section; ‘National Natural Science Foundation of China (6210071334, 62072376)’ has now been corrected to ‘National Natural Science Foundation of China (62102319, 62072376)’.

The three-dimensional genome structure plays a key role in cellular function and gene regulation. Single-cell Hi-C (high-resolution chromosome conformation capture) technology can capture genome structure information at the cell level, which provides the opportunity to study how genome structure varies among different cell types. Recently, a few methods are well designed for single-cell Hi-C clustering. In this manuscript, we perform an in-depth benchmark study of available single-cell Hi-C data clustering methods to implement an evaluation system for multiple clustering frameworks based on both human and mouse datasets. We compare eight methods in terms of visualization and clustering performance. Performance is evaluated using four benchmark metrics including adjusted rand index, normalized mutual information, homogeneity and Fowlkes-Mallows index. Furthermore, we also evaluate the eight methods for the task of separating cells at different stages of the cell cycle based on single-cell Hi-C data.

MOTIVATION: Biomarkers with prognostic ability and biological interpretability can be used to support decision-making in the survival analysis. Genes usually form functional modules to play synergistic roles, such as pathways. Predicting significant features from the functional level can effectively reduce the adverse effects of heterogeneity and obtain more reproducible and interpretable biomarkers. Personalized pathway activation inference can quantify the dysregulation of essential pathways involved in the initiation and progression of cancers, and can contribute to the development of personalized medical treatments. RESULTS: In this study, we propose a novel method to evaluate personalized pathway activation based on signaling entropy for survival analysis (SEPA), which is a new attempt to introduce the information-theoretic entropy in generating pathway representation for each patient. SEPA effectively integrates pathway-level information into gene expression data, converting the high-dimensional gene expression data into the low-dimensional biological pathway activation scores. SEPA shows its classification power on the prognostic pan-cancer genomic data, and the potential pathway markers identified based on SEPA have statistical significance in the discrimination of high- and low-risk cohorts and are likely to be associated with the initiation and progress of cancers. The results show that SEPA scores can be used as an indicator to precisely distinguish cancer patients with different clinical outcomes, and identify important pathway features with strong discriminative power and biological interpretability. AVAILABILITY AND IMPLEMENTATION: The MATLAB-package for SEPA is freely available from https://github.com/xingyili/SEPA. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Abstract How Deep Neural Networks (DNNs) best cope with the understanding of multimedia contents still remains an open problem, mainly due to two factors. First, conventional DNNs cannot effectively learn the representations of the images with sparse visual information. For example, the images describing knowledge concepts in textbooks. Second, existing DNNs cannot effectively capture the fine‐grained interactions between the images and text descriptions. To address these issues, we propose a deep Cross‐Media Grouping Fusion Network (CMGFN), which mainly has two distinctive properties: 1) CMGFN can effectively learn visual features from the images with sparse visual information. This is achieved by first progressively adjusting the attention of convolution filters to valuable visual regions, and then enhancing the use of key visual information in feature construction. 2) By a cross‐media grouping co‐attention mechanism, CMGFN can effectively use the interactions between visual features of different semantics and textual descriptions, to learn cross‐media features representing different fine‐grained semantics in different groups. Empirical studies demonstrate that CMGFN not only achieves state‐of‐the‐art performance on the multimedia documents containing sparse visual information, but also shows superior general applicability on other multimedia data, e.g., the multimedia fake news.

The occurrence and development of cancer is a complex biological process, which usually requires further research by constructing a network. The existing methods of constructing networks are usually based on similarity to construct co-expression networks for analysis, but this similarity-based method usually ignores the data whose expression level increases or decreases at the same time in different groups. Gene pairs may also be related to the evolution of cancer, and that may cause part of the information to be lost. Therefore, we propose a new method of constructing a network based on regression residuals. Then, the module is identified based on the regression residual network according to different stages of cancer. A module network between cancer stages is obtained through the similarity between the modules. Finally, based on the size and density of the subnets in the modular network, enjoy the screening, and finally selected 3 subnets for functional analysis and Veen diagram analysis, and found some biological functions and genes that have been confirmed to be related to cancer.

Tumor-liltrating lymphocytes (TILs) are predictive for response to neoadjuvant treatment in tumors. Still, the abundance of tumor-liltrating cell types has not yet been produced in large quantities, hampering researchers exploring their characteristics. As the levels of genomics or transcriptomics could reflect changes in cell-type proportions, several computational tools have been developed to estimate cell-type abundances based on the reference gene expression proliles. Differential expression analysis is the most widely used to recognize marker genes. However, it ignores the correlation between genes. To this end, we propose a feature selection method, dubbed Discriminative Concrete Autoencoder (DCAE), to identify informative genes on single-cell RNA-seq data, which are then used to quantity cell-type proportions. To evaluate the performance of DCAE on selecting discriminative genes, we conduct experiments on our collected and processed single-cell RNA-seq dataset. First, we compare DCAE to the original Concrete Autoencoder by the cell-type classification accuracies resulting from their selected genes. Then we infer cell-type abundance by using deconvolution function with the chosen small cohort of genes. Next, we evaluate the deconvolution accuracy by the Pearson correlation coefficient between the estimated cell-type proportions and the true proportions, and the corresponding P-value. Finally, we compare the effects of the selected genes and the differential expression genes on the deconvolution accuracy. The results show that our selected genes by DCAE have higher discriminant power to distinguish cell types and effectively infer cell-type abundance. Thus, DCAE provides insights into acquiring candidate biomarkers for cell-type quantification.

Hepatoma is the most common type of primary liver cancer with a high mortality rate in the world. The genetic causes of the disease pathology remain largely unknown. Effective discovery of the genes associated with hepatoma has become important in disease prevention, early diagnosis, and therapeutic treatments. With the developments of molecular networks, graph-based methods have been tremendously successful in predicting disease genes based on the hypothesis of guilt-by-association. Network representation learning (NRL) techniques have accelerated disease gene discovery in recent years because of their powerful network feature extraction ability. However, the current network representation learning-based methods for disease gene discovery did not consider the gene features derived from gene ontology annotations, which apriori group genes with similar functions. To fill this gap, here we propose a novel framework to predict hepatoma-related genes based on representation learning from both protein-protein interactions (PPI) network and gene ontology annotations. Our framework has three steps: learning features from PPI network and gene ontologies using NRL techniques, integrating different features based on autoencoder, predicting hepatoma-related genes using machine learning classifiers. Experiments have demonstrated that our framework could accurately predict hepatoma-related genes with AUROC and AUPRC reaching 0.93 and 0.94, respectively. Compared with other methods using only single representation features, our framework also shows superior performance on hepatoma gene prediction.

Student knowledge diagnosis (SKD) is a fundamental and crucial task in educational data mining (EDM). SKD aims to diagnose student latent knowledge which is inferred from student’s performance. The model used for SKD in EDM comes from two sources: variant classical psychometric approaches, and research on machine learning-based approaches. Tradition psychometric models and their variants diagnosis student knowledge state relying on the question-concept matrix (Q-matrix) empirically designed by experts. However, the expert concepts are expensive and inter-overlapping in their constructions, leading to ambiguous explanations. The recent model Meta-knowledge Dictionary Learning (MetaDL), a learning-based model, proposes a linear sparse dictionary method to mine Q-matrix without expert definition and student latent knowledge representation. MetaDL aims to learn a meta-knowledge dictionary from student responses, where any knowledge entity is a linear combination of a few atoms in the meta-knowledge dictionary. However, a linear model cannot capture complex features from the student learning process and MetaDL fails to solve the missing data. This paper proposes a novel Variational Student Knowledge Diagnosis (VarSKD) method that extends the linear sparse representation of student latent knowledge space into non-linear probabilistic sparse representation. This model based on variational sparse coding can obtain better student latent knowledge representation. Furthermore, extensive experimental results on real-world datasets demonstrate the prediction accuracy and effective power of VarSKD framework.

Prediction of undergraduate grades before their course enrollments is beneficial to the student’s learning plan on selective courses and failure warnings to compulsory courses in Chinese higher education. This study proposed to use a deep learning-based model composed of sparse attention layers, convolutional neural layers, and a fully connected layer, called Sparse Attention Convolutional Neural Networks (SACNN), to predict undergraduate grades. Concretely, sparse attention layers response to the fact that courses have different contributions to the grade prediction of the target course; convolutional neural layers aim to capture the one-dimensional temporal feature on these courses organized in terms; the fully connected layer is to complete the final classification based on achieved features. We collected a dataset including grade records, student’s demographics and course descriptions from our institution in the past five years. The dataset contained about 54k grade records from 1307 students and 137 courses, where all mentioned methods were evaluated by the hold-out evaluation. The result shows SACNN achieves 81% prediction precision and 85% accuracy on the failure prediction, which is more effective than those compared methods. Besides, SACNN delivers a potential explanation to the reason of the predicted result, thanks to the sparse attention layer. This study provides a useful technique for personalized learning and course relationship discovery in undergraduate education.

Different subtypes of the same cancer often show distinct genomic signatures and require targeted treatments. The differences at the cellular and molecular levels of tumor microenvironment in different cancer subtypes have significant effects on tumor pathogenesis and prognostic outcomes. Although there have been significant researches on the prognostic association of tumor infiltrating lymphocytes in selected histological subtypes, few investigations have systemically reported the prognostic impacts of immune cells in molecular subtypes, as quantified by machine learning approaches on multi-omics datasets. This paper describes a new computational framework, ProTICS, to quantify the differences in the proportion of immune cells in tumor microenvironment and estimate their prognostic effects in different subtypes. First, we stratified patients into molecular subtypes based on gene expression and methylation profiles by applying nonnegative tensor factorization technique. Then we quantified the proportion of cell types in each specimen using an mRNA-based deconvolution method. For tumors in each subtype, we estimated the prognostic effects of immune cell types by applying Cox proportional hazard regression. At the molecular level, we also predicted the prognosis of signature genes for each subtype. Finally, we benchmarked the performance of ProTICS on three TCGA datasets and another independent METABRIC dataset. ProTICS successfully stratified tumors into different molecular subtypes manifested by distinct overall survival. Furthermore, the different immune cell types showed distinct prognostic patterns with respect to molecular subtypes. This study provides new insights into the prognostic association between immune cells and molecular subtypes, showing the utility of immune cells as potential prognostic markers. Availability: R code is available at https://github.com/liu-shuhui/ProTICS.

Accurately identifying potential drug-target interactions (DTIs) is a key step in drug discovery. Although many related experimental studies have been carried out for identifying DTIs in the past few decades, the biological experiment-based DTI identification is still timeconsuming and expensive. Therefore, it is of great significance to develop effective computational methods for identifying DTIs. In this paper, we develop a novel 'end-to-end' learning-based framework based on heterogeneous 'graph' convolutional networks for 'DTI' prediction called end-to-end graph (EEG)-DTI. Given a heterogeneous network containing multiple types of biological entities (i.e. drug, protein, disease, side-effect), EEG-DTI learns the low-dimensional feature representation of drugs and targets using a graph convolutional networks-based model and predicts DTIs based on the learned features. During the training process, EEG-DTI learns the feature representation of nodes in an end-to-end mode. The evaluation test shows that EEG-DTI performs better than existing state-of-art methods. The data and source code are available at: https://github.com/MedicineBiology-AI/EEG-DTI.

MOTIVATION: The emergence of abundant biological networks, which benefit from the development of advanced high-throughput techniques, contributes to describing and modeling complex internal interactions among biological entities such as genes and proteins. Multiple networks provide rich information for inferring the function of genes or proteins. To extract functional patterns of genes based on multiple heterogeneous networks, network embedding-based methods, aiming to capture non-linear and low-dimensional feature representation based on network biology, have recently achieved remarkable performance in gene function prediction. However, existing methods do not consider the shared information among different networks during the feature learning process. RESULTS: Taking the correlation among the networks into account, we design a novel semi-supervised autoencoder method to integrate multiple networks and generate a low-dimensional feature representation. Then we utilize a convolutional neural network based on the integrated feature embedding to annotate unlabeled gene functions. We test our method on both yeast and human datasets and compare with three state-of-the-art methods. The results demonstrate the superior performance of our method. We not only provide a comprehensive analysis of the performance of the newly proposed algorithm but also provide a tool for extracting features of genes based on multiple networks, which can be used in the downstream machine learning task. AVAILABILITY: DeepMNE-CNN is freely available at https://github.com/xuehansheng/DeepMNE-CNN. CONTACT: [email protected]; [email protected]; [email protected].

BACKGROUND: It is important to understand the composition of cell type and its proportion in intact tissues, as changes in certain cell types are the underlying cause of disease in humans. Although compositions of cell type and ratios can be obtained by single-cell sequencing, single-cell sequencing is currently expensive and cannot be applied in clinical studies involving a large number of subjects. Therefore, it is useful to apply the bulk RNA-Seq dataset and the single-cell RNA dataset to deconvolute and obtain the cell type composition in the tissue. RESULTS: By analyzing the existing cell population prediction methods, we found that most of the existing methods need the cell-type-specific gene expression profile as the input of the signature matrix. However, in real applications, it is not always possible to find an available signature matrix. To solve this problem, we proposed a novel method, named DCap, to predict cell abundance. DCap is a deconvolution method based on non-negative least squares. DCap considers the weight resulting from measurement noise of bulk RNA-seq and calculation error of single-cell RNA-seq data, during the calculation process of non-negative least squares and performs the weighted iterative calculation based on least squares. By weighting the bulk tissue gene expression matrix and single-cell gene expression matrix, DCap minimizes the measurement error of bulk RNA-Seq and also reduces errors resulting from differences in the number of expressed genes in the same type of cells in different samples. Evaluation test shows that DCap performs better in cell type abundance prediction than existing methods. CONCLUSION: DCap solves the deconvolution problem using weighted non-negative least squares to predict cell type abundance in tissues. DCap has better prediction results and does not need to prepare a signature matrix that gives the cell-type-specific gene expression profile in advance. By using DCap, we can better study the changes in cell proportion in diseased tissues and provide more information on the follow-up treatment of diseases.

BACKGROUND: Single-cell RNA sequencing (scRNA-seq) enables the possibility of many in-depth transcriptomic analyses at a single-cell resolution. It's already widely used for exploring the dynamic development process of life, studying the gene regulation mechanism, and discovering new cell types. However, the low RNA capture rate, which cause highly sparse expression with dropout, makes it difficult to do downstream analyses. RESULTS: We propose a new method SCC to impute the dropouts of scRNA-seq data. Experiment results show that SCC gives competitive results compared to two existing methods while showing superiority in reducing the intra-class distance of cells and improving the clustering accuracy in both simulation and real data. CONCLUSIONS: SCC is an effective tool to resolve the dropout noise in scRNA-seq data. The code is freely accessible at https://github.com/nwpuzhengyan/SCC .

One type of cancer usually consists of several subtypes with distinct clinical implications, thus the cancer subtype prediction is an important task in disease diagnosis and therapy. Utilizing one type of data from molecular layers in biological system to predict is difficult to bridge the cancer genome to cancer phenotypes, since the genome is neither simple nor independent but rather complicated and dysregulated from multiple molecular mechanisms. Similarity Network Fusion (SNF) has been recently proposed to integrate diverse omics data for improving the understanding of tumorigenesis. SNF adopts Euclidean distance to measure the similarity between patients, which shows some limitations. In this article, we introduce a novel prediction technique as an extension of SNF, namely Deep Subspace Fusion Clustering (DSFC). DSFC utilizes auto-encoder and data self-expressiveness approaches to guide a deep subspace model, which can achieve effective expression of discriminative similarity between patients. As a result, the dissimilarity between inter-cluster is delivered and enhanced compactness of intra-cluster is achieved at the same time. The validity of DSFC is examined by extensive simulations over six different cancer through three levels omics data. The survival analysis demonstrates that DSFC delivers comparable or even better results than many state-of-the-art integrative methods.

MOTIVATION: The rapid improvement of phenotyping capability, accuracy and throughput have greatly increased the volume and diversity of phenomics data. A remaining challenge is an efficient way to identify phenotypic patterns to improve our understanding of the quantitative variation of complex phenotypes, and to attribute gene functions. To address this challenge, we developed a new algorithm to identify emerging phenomena from large-scale temporal plant phenotyping experiments. An emerging phenomenon is defined as a group of genotypes who exhibit a coherent phenotype pattern during a relatively short time. Emerging phenomena are highly transient and diverse, and are dependent in complex ways on both environmental conditions and development. Identifying emerging phenomena may help biologists to examine potential relationships among phenotypes and genotypes in a genetically diverse population and to associate such relationships with the change of environments or development. RESULTS: We present an emerging phenomenon identification tool called Temporal Emerging Phenomenon Finder (TEP-Finder). Using large-scale longitudinal phenomics data as input, TEP-Finder first encodes the complicated phenotypic patterns into a dynamic phenotype network. Then, emerging phenomena in different temporal scales are identified from dynamic phenotype network using a maximal clique based approach. Meanwhile, a directed acyclic network of emerging phenomena is composed to model the relationships among the emerging phenomena. The experiment that compares TEP-Finder with two state-of-art algorithms shows that the emerging phenomena identified by TEP-Finder are more functionally specific, robust and biologically significant. AVAILABILITY AND IMPLEMENTATION: The source code, manual and sample data of TEP-Finder are all available at: http://phenomics.uky.edu/TEP-Finder/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Drug-target interaction prediction is of great significance for narrowing down the scope of candidate medications, and thus is a vital step in drug discovery. Because of the particularity of biochemical experiments, the development of new drugs is not only costly, but also time-consuming. Therefore, the computational prediction of drug target interactions has become an essential way in the process of drug discovery, aiming to greatly reducing the experimental cost and time. RESULTS: We propose a learning-based method based on feature representation learning and deep neural network named DTI-CNN to predict the drug-target interactions. We first extract the relevant features of drugs and proteins from heterogeneous networks by using the Jaccard similarity coefficient and restart random walk model. Then, we adopt a denoising autoencoder model to reduce the dimension and identify the essential features. Third, based on the features obtained from last step, we constructed a convolutional neural network model to predict the interaction between drugs and proteins. The evaluation results show that the average AUROC score and AUPR score of DTI-CNN were 0.9416 and 0.9499, which obtains better performance than the other three existing state-of-the-art methods. CONCLUSIONS: All the experimental results show that the performance of DTI-CNN is better than that of the three existing methods and the proposed method is appropriately designed.

BACKGROUND: Single-cell RNA sequencing (scRNA-seq) provides an effective tool to investigate the transcriptomic characteristics at the single-cell resolution. Due to the low amounts of transcripts in single cells and the technical biases in experiments, the raw scRNA-seq data usually includes large noise and makes the downstream analyses complicated. Although many methods have been proposed to impute the noisy scRNA-seq data in recent years, few of them take into account the prior associations across genes in imputation and integrate multiple types of imputation data to identify cell types. RESULTS: We present a new framework, NetImpute, towards the identification of cell types from scRNA-seq data by integrating multiple types of biological networks. We employ a statistic method to detect the noise data items in scRNA-seq data and develop a new imputation model to estimate the real values of data noise by integrating the PPI network and gene pathways. Meanwhile, based on the data imputed by multiple types of biological networks, we propose an integrated approach to identify cell types from scRNA-seq data. Comprehensive experiments demonstrate that the proposed network-based imputation model can estimate the real values of noise data items accurately and integrating the imputation data based on multiple types of biological networks can improve the identification of cell types from scRNA-seq data. CONCLUSIONS: Incorporating the prior gene associations in biological networks can potentially help to improve the imputation of noisy scRNA-seq data and integrating multiple types of network-based imputation data can enhance the identification of cell types. The proposed NetImpute provides an open framework for incorporating multiple types of biological network data to identify cell types from scRNA-seq data.

BACKGROUND: Network alignment is an efficient computational framework in the prediction of protein function and phylogenetic relationships in systems biology. However, most of existing alignment methods focus on aligning PPIs based on static network model, which are actually dynamic in real-world systems. The dynamic characteristic of PPI networks is essential for understanding the evolution and regulation mechanism at the molecular level and there is still much room to improve the alignment quality in dynamic networks. RESULTS: In this paper, we proposed a novel alignment algorithm, Twadn, to align dynamic PPI networks based on a strategy of time warping. We compare Twadn with the existing dynamic network alignment algorithm DynaMAGNA++ and DynaWAVE and use area under the receiver operating characteristic curve and area under the precision-recall curve as evaluation indicators. The experimental results show that Twadn is superior to DynaMAGNA++ and DynaWAVE. In addition, we use protein interaction network of Drosophila to compare Twadn and the static network alignment algorithm NetCoffee2 and experimental results show that Twadn is able to capture timing information compared to NetCoffee2. CONCLUSIONS: Twadn is a versatile and efficient alignment tool that can be applied to dynamic network. Hopefully, its application can benefit the research community in the fields of molecular function and evolution.

Multi-networks integration methods have achieved prominent performance on many network-based tasks, but these approaches often incur information loss problem. In this paper, we propose a novel multi-networks representation learning method based on semi-supervised autoencoder, termed as DeepMNE, which captures complex topological structures of each network and takes the correlation among multinetworks into account. The experimental results on two realworld datasets indicate that DeepMNE outperforms the existing state-of-the-art algorithms.

Colorectal cancer (CRC) is one of the malignant tumors with high morbidity and mortality. A prevalent method for studying colorectal cancer is to identify differentially expressed genes (DEGs) between control and patient samples, followed by the pathway enrichment analyses. However, many of those studies ignore the fact that different pathological stages of the cancer are often highly different from each other. The mixture of those heterogeneous samples may lack the efficiency of identifying the real DEGs, and loss the opportunity to analyze the dynamic evolution process of cancer. In this study, we develop a feasible framework to identify function evolution processes of cancers according to their pathological stages. Firstly, the limma package was used to identify DEG sets between control and CRC stage I, II, III, and IV samples, separately. Secondly, a pathway interaction network was constructed by taking a comprehensive analysis of pathways at individual stages, and a functional module interaction network was also generated independently by clustering genes into modules in a PPI network. The relationship between pathways and modules in adjacent stages was analyzed for all stages of CRC. A total of 479, 313, 349, and 383 DEGs were identified and they were enriched in 17, 16, 20, and 24 pathways, respectively. A functional evolution network was constructed by using those modules, and two significant evolution processes a2-b1-c2-d1 (Mod1) and a1-b2-c1-d2 (Mod2) were identified which may play critical roles in the development of CRC. The framework proposed in this study can be used to explore molecular mechanisms and evolution processes of CRC.

It is important to understand the cell-type composition and its proportion in tissues. Some previous experiments have shown that the variation of gene expression in certain cell types may lead to disease. Although cell type composition and proportion can be obtained by single-cell RNA-sequencing (scRNA-sq), using scRNA-sq is expensive and cannot be applied in clinical studies involving a large number of subjects currently. Therefore, it is urgent to develop a method to deconvolute the Bulk RNA-Seq data to obtain the cell type composition in the tissue. Most of the existing methods require the signature matrix, which provides the cell type-specific gene expression profile, as input. However, the signature matrix is not always available for some types of tissue, and it is not always possible to find a suitable cell type-specific gene expression profile. To solve this problem, we propose a novel method, named DCap, to predict cell abundance. Different from non-negative least squares, DCap performs weighted iterative calculation based on least squares. By weighting bulk tissue gene expression matrix and single-cell gene expression matrix, DCap minimizes the measurement error of Bulk RNA-Seq and error resulting from the difference in the amount of genes in the same cell type among different samples. DCap solves the deconvolution problem by using weighted non-negative least squares to predict cell type abundance. DCap does not need to prepare a suitable signature matrix in advance, and the evaluation test shows that DCap performs better in cell type abundance prediction than existing methods.

Single-cell RNA sequencing (scRNAseq) can reveal biological diversity at the cellular level that are unexplored by bulk RNA sequencing (RNAseq), but they suffer from the excessive zero expression counts and the limitation of the scalability in practice. Here, we propose a non-linear generative autoencoder based method, scSVA, relying on an integration of variational autoencoder and dropout imputations. Specifically, scSVA automatically identifies the dropouts and recovery these values only to avoid introducing new biases. Then, scSVA utilizes stochastic optimization and deep neural network to extract the low-dimensional embedding from gene expression levels. We illustrate the benefits of scSVA through in-depth real analyses of six published scRNAseq data sets. scSVA is up to 1.3 times more powerful cell clustering accuracy than existing approaches. The high power of scSVA allows us to identify new cell types that reveal new biology from scRNAseq data that otherwise cannot be revealed by existing approaches.

As more and more single-cell RNA-seq (scRNA-seq) datasets become available, carrying out compare between them is key. However, this task is challengeable due to differences caused by different experiment. We proposed a single cell alignment method using deep autoencoder followed by k-nearst-neighbor cells (scadKNN), which learns the feature representation of the data while eliminating batch effects and dropouts through deep autoencoder and uses the low-dimensional feature to align cell types, thereby reducing calculation effort and improving alignment accuracy. Experiments using different real datasets are employed to showcase the effectiveness of the proposed approach.

Node Embedding, which uses low-dimensional non-linear feature vectors to represent nodes in the network, has shown a great promise, not only because it is easy-to-use for downstream tasks, but also because it has achieved great success on many network analysis tasks. One of the challenges has been how to develop a node embedding method for integrating topological information from multiple networks. To address this critical problem, we propose a novel node embedding, called DeepMNE, for multi-network integration using a deep semi-supervised autoencoder. The key point of DeepMNE is that it captures complex topological structures of multiple networks and utilizes correlation among multiple networks as constraints. We evaluate DeepMNE in node classification task and link prediction task on four real-world datasets. The experimental results demonstrate that DeepMNE shows superior performance over seven state-of-the-art single-network and multi-network embedding algorithms.

MOTIVATION: A microRNA (miRNA) is a type of non-coding RNA, which plays important roles in many biological processes. Lots of studies have shown that miRNAs are implicated in human diseases, indicating that miRNAs might be potential biomarkers for various types of diseases. Therefore, it is important to reveal the relationships between miRNAs and diseases/phenotypes. RESULTS: We propose a novel learning-based framework, MDA-CNN, for miRNA-disease association identification. The model first captures interaction features between diseases and miRNAs based on a three-layer network including disease similarity network, miRNA similarity network and protein-protein interaction network. Then, it employs an auto-encoder to identify the essential feature combination for each pair of miRNA and disease automatically. Finally, taking the reduced feature representation as input, it uses a convolutional neural network to predict the final label. The evaluation results show that the proposed framework outperforms some state-of-the-art approaches in a large margin on both tasks of miRNA-disease association prediction and miRNA-phenotype association prediction. AVAILABILITY AND IMPLEMENTATION: The source code and data are available at https://github.com/Issingjessica/MDA-CNN. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: The mechanism of many complex diseases has not been detected accurately in terms of their stage evolution. Previous studies mainly focus on the identification of associations between genes and individual diseases, but less is known about their associations with specific disease stages. Exploring biological modules through different disease stages could provide valuable knowledge to genomic and clinical research. RESULTS: In this study, we proposed a powerful and versatile framework to identify stage-specific cancer related genes and their dynamic modules by integrating multiple datasets. The discovered modules and their specific-signature genes were significantly enriched in many relevant known pathways. To further illustrate the dynamic evolution of these clinical-stages, a pathway network was built by taking individual pathways as vertices and the overlapping relationship between their annotated genes as edges. CONCLUSIONS: The identified pathway network not only help us to understand the functional evolution of complex diseases, but also useful for clinical management to select the optimum treatment regimens and the appropriate drugs for patients.

BACKGROUNDS: There is evidence to suggest that lncRNAs are associated with distinct and diverse biological processes. The dysfunction or mutation of lncRNAs are implicated in a wide range of diseases. An accurate computational model can benefit the diagnosis of diseases and help us to gain a better understanding of the molecular mechanism. Although many related algorithms have been proposed, there is still much room to improve the accuracy of the algorithm. RESULTS: We developed a novel algorithm, BiWalkLDA, to predict disease-related lncRNAs in three real datasets, which have 528 lncRNAs, 545 diseases and 1216 interactions in total. To compare performance with other algorithms, the leave-one-out validation test was performed for BiWalkLDA and three other existing algorithms, SIMCLDA, LDAP and LRLSLDA. Additional tests were carefully designed to analyze the parameter effects such as α, β, l and r, which could help user to select the best choice of these parameters in their own application. In a case study of prostate cancer, eight out of the top-ten disease-related lncRNAs reported by BiWalkLDA were previously confirmed in literatures. CONCLUSIONS: In this paper, we develop an algorithm, BiWalkLDA, to predict lncRNA-disease association by using bi-random walks. It constructs a lncRNA-disease network by integrating interaction profile and gene ontology information. Solving cold-start problem by using neighbors' interaction profile information. Then, bi-random walks was applied to three real biological datasets. Results show that our method outperforms other algorithms in predicting lncRNA-disease association in terms of both accuracy and specificity. AVAILABILITY: https://github.com/screamer/BiwalkLDA.

BACKGROUND: Transcription factors (TFs) play important roles in the regulation of gene expression. They can activate or block transcription of downstream genes in a manner of binding to specific genomic sequences. Therefore, motif discovery of these binding preference patterns is of central significance in the understanding of molecular regulation mechanism. Many algorithms have been proposed for the identification of transcription factor binding sites. However, it remains a challengeable problem. RESULTS: Here, we proposed a novel motif discovery algorithm based on support vector machine (MD-SVM) to learn a discriminative model for TF binding sites. MD-SVM firstly obtains position weight matrix (PWM) from a set of training datasets. Then it translates the MD problem into a computational framework of multiple instance learning (MIL). It was applied to several real biological datasets. Results show that our algorithm outperforms MI-SVM in terms of both accuracy and specificity. CONCLUSIONS: In this paper, we modeled the TF motif discovery problem as a MIL optimization problem. The SVM algorithm was adapted to discriminate positive and negative bags of instances. Compared to other svm-based algorithms, MD-SVM show its superiority over its competitors in term of ROC AUC. Hopefully, it could be of benefit to the research community in the understanding of molecular functions of DNA functional elements and transcription factors.

BACKGROUND: The Gene Ontology (GO) knowledgebase is the world's largest source of information on the functions of genes. Since the beginning of GO project, various tools have been developed to perform GO enrichment analysis experiments. GO enrichment analysis has become a commonly used method of gene function analysis. Existing GO enrichment analysis tools do not consider tissue-specific information, although this information is very important to current research. RESULTS: In this paper, we built an easy-to-use web tool called TS-GOEA that allows users to easily perform experiments based on tissue-specific GO enrichment analysis. TS-GOEA uses strict threshold statistical method for GO enrichment analysis, and provides statistical tests to improve the reliability of the analysis results. Meanwhile, TS-GOEA provides tools to compare different experimental results, which is convenient for users to compare the experimental results. To evaluate its performance, we tested the genes associated with platelet disease with TS-GOEA. CONCLUSIONS: TS-GOEA is an effective GO analysis tool with unique features. The experimental results show that our method has better performance and provides a useful supplement for the existing GO enrichment analysis tools. TS-GOEA is available at http://120.77.47.2:5678.

BACKGROUND: Single cell RNA sequencing (scRNA-seq) is applied to assay the individual transcriptomes of large numbers of cells. The gene expression at single-cell level provides an opportunity for better understanding of cell function and new discoveries in biomedical areas. To ensure that the single-cell based gene expression data are interpreted appropriately, it is crucial to develop new computational methods. RESULTS: In this article, we try to re-construct a neural network based on Gene Ontology (GO) for dimension reduction of scRNA-seq data. By integrating GO with both unsupervised and supervised models, two novel methods are proposed, named GOAE (Gene Ontology AutoEncoder) and GONN (Gene Ontology Neural Network) respectively. CONCLUSIONS: The evaluation results show that the proposed models outperform some state-of-the-art dimensionality reduction approaches. Furthermore, incorporating with GO, we provide an opportunity to interpret the underlying biological mechanism behind the neural network-based model.

BACKGROUND: The classification of cancer subtypes is of great importance to cancer disease diagnosis and therapy. Many supervised learning approaches have been applied to cancer subtype classification in the past few years, especially of deep learning based approaches. Recently, the deep forest model has been proposed as an alternative of deep neural networks to learn hyper-representations by using cascade ensemble decision trees. It has been proved that the deep forest model has competitive or even better performance than deep neural networks in some extent. However, the standard deep forest model may face overfitting and ensemble diversity challenges when dealing with small sample size and high-dimensional biology data. RESULTS: In this paper, we propose a deep learning model, so-called BCDForest, to address cancer subtype classification on small-scale biology datasets, which can be viewed as a modification of the standard deep forest model. The BCDForest distinguishes from the standard deep forest model with the following two main contributions: First, a named multi-class-grained scanning method is proposed to train multiple binary classifiers to encourage diversity of ensemble. Meanwhile, the fitting quality of each classifier is considered in representation learning. Second, we propose a boosting strategy to emphasize more important features in cascade forests, thus to propagate the benefits of discriminative features among cascade layers to improve the classification performance. Systematic comparison experiments on both microarray and RNA-Seq gene expression datasets demonstrate that our method consistently outperforms the state-of-the-art methods in application of cancer subtype classification. CONCLUSIONS: The multi-class-grained scanning and boosting strategy in our model provide an effective solution to ease the overfitting challenge and improve the robustness of deep forest model working on small-scale data. Our model provides a useful approach to the classification of cancer subtypes by using deep learning on high-dimensional and small-scale biology data.

BACKGROUND: The discovery of functionally conserved proteins is a tough and important task in system biology. Global network alignment provides a systematic framework to search for these proteins from multiple protein-protein interaction (PPI) networks. Although there exist many web servers for network alignment, no one allows to perform global multiple network alignment tasks on users' test datasets. RESULTS: Here, we developed a web server WebNetcoffee based on the algorithm of NetCoffee to search for a global network alignment from multiple networks. To build a series of online test datasets, we manually collected 218,339 proteins, 4,009,541 interactions and many other associated protein annotations from several public databases. All these datasets and alignment results are available for download, which can support users to perform algorithm comparison and downstream analyses. CONCLUSION: WebNetCoffee provides a versatile, interactive and user-friendly interface for easily running alignment tasks on both online datasets and users' test datasets, managing submitted jobs and visualizing the alignment results through a web browser. Additionally, our web server also facilitates graphical visualization of induced subnetworks for a given protein and its neighborhood. To the best of our knowledge, it is the first web server that facilitates the performing of global alignment for multiple PPI networks. AVAILABILITY: http://www.nwpu-bioinformatics.com/WebNetCoffee.

BACKGROUND: Recently, measuring phenotype similarity began to play an important role in disease diagnosis. Researchers have begun to pay attention to develop phenotype similarity measurement. However, existing methods ignore the interactions between phenotype-associated proteins, which may lead to inaccurate phenotype similarity. RESULTS: We proposed a network-based method PhenoNet to calculate the similarity between phenotypes. We localized phenotypes in the network and calculated the similarity between phenotype-associated modules by modeling both the inter- and intra-similarity. CONCLUSIONS: PhenoNet was evaluated on two independent evaluation datasets: gene ontology and gene expression data. The result shows that PhenoNet performs better than the state-of-art methods on all evaluation tests.

The classification of cancer subtypes is of great importance in cancer disease diagnosis and therapy. Many supervised learning methods have been applied to classification of cancer subtypes in the past few years, especially of deep learning based methods. Recently, a deep forest model has been proposed as an alternative of deep neural networks to learn hyper-representations by using cascade ensemble decision trees, and it has been proved that deep forest model has competitive or even better performance than deep neural networks. However, the original deep forest may face under-fitting and ensemble diversity problems when dealing with small sample size, and high-dimension biology data. It is important to improve the deep forest model to work better on small-scale biology data. In this paper, we propose a deep learning model to follow the mission of cancer subtype classification on small-scale biology data sets, which can be viewed as modification of original deep forest model. Our model distinguishes from the original deep forest model with two main contributions: First, a named multi-class-scanning method is proposed to train multiple simple binary classifiers to encourage diversity of ensemble. Meanwhile, the fitting quality of each classifier is considered in representations learning. Second, we propose a boosting strategy to emphasize more important features in cascade forests of representations learning, thus to propagate the benefits of discriminative features among layers to improve the overall classification performance. Systematical experiments on both microarray and RNA-seq data sets demonstrate that our method consistently outperforms the most state-of-the-art classification methods in application of cancer subtype classifications.

Miniature inverted-repeat transposable element (M ITE) is a type of class II non-autonomous transposable element playing a crucial role in the process of evolution in biology. Development of bioinformatics tools that are capable of effectively identifying MITEs can enable genome-wide studies of MITE patterns in eukaryotes. Here, we present a fast, accurate and memory-efficient tool, MiteFinder, for the identification of MITEs from genomics sequences. MiteFinder distinguishes itself from other existing methods by building k-mer indexes for large genomes, which can effectively identify all possible inverted repeats. To calculate the likelihood of how much a given sequence is a MITE, we introduce a log-ratio scoring model that uses the distribution of sequence pattern in two models, mite model (M) and null model (N). The results suggest that MiteFinder outperforms existing tools in both precision and recall. Besides, it is much faster and more memory-efficient than other tools in the detection. The source code is freely accessible at the website: https://github.com/screamer/miteFinder.

Recently, measuring phenotype similarity began to play an important role in disease diagnosis. Researchers have begun to pay attention to develop phenotype similarity measurement. However, existing methods ignore the interactions between phenotype-associated proteins, which may lead to inaccurate phenotype similarity. We proposed a network-based method PhenoNet to calculate the similarity between phenotypes. We localized phenotypes in the network and calculated the similarity between phenotype-associated modules by modeling both the inter- and intra-similarity. PhenoNet was evaluated on two independent evaluation datasets: gene ontology and gene expression data. The result shows that PhenoNet performs better than the state-of-art methods on all evaluation tests.

BACKGROUND: The Gene Ontology (GO) is a community-based bioinformatics resource that employs ontologies to represent biological knowledge and describes information about gene and gene product function. GO includes three independent categories: molecular function, biological process and cellular component. For better biological reasoning, identifying the biological relationships between terms in different categories are important. However, the existing measurements to calculate similarity between terms in different categories are either developed by using the GO data only or only take part of combined gene co-function network information. RESULTS: We propose an iterative ranking-based method called C r o G O2 to measure the cross-categories GO term similarities by incorporating level information of GO terms with both direct and indirect interactions in the gene co-function network. CONCLUSIONS: The evaluation test shows that C r o G O2 performs better than the existing methods. A genome-specific term association network for yeast is also generated by connecting terms with the high confidence score. The linkages in the term association network could be supported by the literature. Given a gene set, the related terms identified by using the association network have overlap with the related terms identified by GO enrichment analysis.

Although disease diagnosis have greatly benefited from next generation sequencing technologies, it is still difficult to make the right diagnosis based on purely sequencing technologies for many diseases with complex phenotypes and high genetic heterogeneity. Recently, calculating Human Phenotype Ontology (HPO)-based phenotype semantic similarity has contributed a lot for completing disease diagnosis. However, factors which affect the accuracy of HPO-based semantic similarity have not been evaluated systematically. In this study, we propose a new framework called HPOFactor to evaluate these factors.

It is critical yet remains to be challenging to make right disease diagnosis based on complex clinical characteristic and heterogeneous genetic background. Recently, Human Phenotype Ontology (HPO)-based phenotype similarity has been widely used to aid disease diagnosis. However, the existing measurements are revised based on the Gene Ontology-based term similarity models, which are not optimized for human phenotype ontologies. We propose a new similarity measure called PhenoSim. Our model includes a noise reduction component to model the noisy patient phenotype data, and a path-constrained Information Content-based method for measuring phenotype semantics similarity. Evaluation tests showed that PhenoSim could improve the performance of HPO-based phenotype similarity measurement.

There is an urgent need to discover or predict DDIs, which would cause serious adverse drug reactions. However, preclinical detection of DDIs bear high cost. Similarity-based computational approaches can be the assistance of experimental approaches. Utilizing pre-market drug similarities, they are able to predict DDIs on a large scale. However, they neglect the topological structure among DDIs and non-DDIs and have a burden of slow training and much memory. Or, they bear the bias that the pairs between a newly-given drug and the drugs having many DDIs tend to obtain high ranks. More importantly, they lack an effective combination of multiple predictions. To address these issues, we develop a local classification-based model (LCM), which has the advantages of faster training, less memory requirement as well as no that bias. We further design a novel supervised algorithm of fusion based on Dempster-Shafer (DS) theory of evidence for combine multiple predictions. Finally, the experiments demonstrate that our LCM-DS is significantly superior to three state-of-the-art approaches and outperforms both individual LCMs and classical fusion algorithms.

The identification of cancer-related genes is important towards the understanding of complex genetic diseases. Although many machine learning algorithms are proposed to identify disease-related genes, they often either have poor performance to identify locus heterogeneity cancer-related genes or are not applicable to predict individual-disease-related genes due to the lack of positive instances (imbalanced classification). To overcome these two issues, a two-step logistic regression (LR) based algorithm is proposed in this study for identifying individual-cancer-related genes. A set of high potential cancer-class-related genes is first generated in step 1, followed by a second round of LR-based algorithm conducted on this smaller dataset for identifying individual-cancer-related genes. Numerical experiments show that the proposed two-step LR-based algorithm not only works well for locus heterogeneity data, but also has good performance to handle the imbalanced classification problem. The individual-cancer-related gene identification experiments achieve AUC values of around 0.85 when the threshold of posterior probability is chosen between 0.3 and 0.6. All evaluations are conducted by using the leave-one-out cross validation method.

Mining the protein complexes and functional modules from protein-protein interaction (PPI) networks is vital to understand the mechanism of cellular components and protein functions. Most of the proposed methods had solely focused on static properties of the PPI networks since the available PPI data are static. However, cellular systems are highly dynamic. That is, the interactions of proteins are responsive to environmental cues to accomplish diverse cellular functions. It is important to consider the dynamic inherent within the PPI networks to identify protein complexes and functional modules. In addition, most computational methods did not distinguish between protein complexes and functional modules. It is important to distinguish between them since they are different protein organizations. In this paper, we propose a novel framework to analyze the PPI networks in dynamic conditions by integrating time-series gene expression profiles data and subcelluar localization data. The algorithm, CBMI, is developed to identify protein complexes in integrated PPI networks. By investigating multiple perspectives of proteins in the PPI networks, we identify the “dynamic” hubs in the PPI networks, and then present a new method to discover the functional modules in the PPI networks. The experimental results show that the integration of temporal gene expression data and subcelluar localization data with PPI data contributes to extracting the protein complexes more precisely. Comprehensive evaluations based on f-measure and functional annotations in MIPS database reveal that our algorithm, CBMI, outperforms other previous algorithms in identifying protein complexes, and the detected functional modules are statistically significant in terms of functional annotations. The proposed framework provides a new clue to distinguish between protein complexes and functional modules, and the developed algorithms can be an effective technique for the identification of them.