CircRNA-miRNA interaction (CMI) plays a pivotal role in disease therapeutics and drug discovery. However, existing methods face several challenges in modeling complex biological networks and zero-shot learning scenarios. Biological networks encapsulate rich biological information, yet current approaches often fail to fully exploit this depth. Moreover, zero-shot prediction requires models to identify new interactions without relying on previously observed samples, imposing stringent requirements on generalization capabilities. To address these limitations, we propose a dual-channel learning framework leveraging State space modeling for Zero-shot CMI prediction (ZeroStem). ZeroStem first enhances the biological relevance of node using prior knowledge, and employs a graph Transformer to extract macro-topological representations. Subsequently, it generates semantic subgraphs based on meta-paths to focus on specific biological relationships, utilizing the Mamba to extract micro-semantic representations via state space modeling. Finally, macro-topological and micro-semantic representations are seamlessly integrated through linear transformation and residual connections, enabling high-precision zero-shot CMI prediction. Extensive experiments on multiple benchmark datasets demonstrate that ZeroStem significantly outperforms existing methods, validating its efficiency and robust generalization in CMI prediction. Case studies further illustrate that ZeroStem offers novel insights into the molecular mechanisms underlying intricate disease-associated networks.

BACKGROUND: Organoid image segmentation is essential for quantitative analysis in disease modeling and drug screening, yet remains highly challenging due to substantial morphological variability and blurred boundaries in organoid images. Existing approaches often struggle to achieve a favorable balance between segmentation accuracy and computational efficiency. RESULTS: In this paper, DeShiftNet, a lightweight segmentation framework, is proposed to extract discriminative features with high accuracy while maintaining low computational overhead. The model incorporates a deformable-shifted encoding strategy that adaptively samples local structures. It also includes a cross-attention-guided decoder for selective multi-scale feature alignment. Furthermore, a deformable multi-scale contextual refinement module enhances boundary coherence and contextual consistency. Extensive experiments on the multi-type OrganoID dataset show that DeShiftNet achieves competitive performance compared with recent segmentation models, while maintaining only 1.78M parameters and 2.65 GFLOPs. Notably, DeShiftNet achieves a Dice score of 0.961 on the Lung subset. CONCLUSION: These results indicate its potential practical value for efficient organoid segmentation in high-throughput experimental workflows.

Accurate and automated identification of organoids from bright-field images is essential for enabling high-throughput drug screening and precision medicine. Organoids, as 3-D in vitro cellular models, closely recapitulate the functional and structural characteristics of their tissue or organ of origin, presenting an unprecedented opportunity for biomedical research. However, the complexity of bright-field microscopy images, including heterogeneous backgrounds and diverse organoid morphologies, poses significant challenges for existing computational methods, often hindering robust feature extraction and high-throughput analysis. To address these issues at the intersection of computational vision and organoid biology, we propose FEMSSorg, a novel organoid recognition framework designed to adaptively aggregate multiscale scan-selected state space features through a fuzzy mixture-of-experts (FuzzyMoE) scoring mechanism. FEMSSorg introduces a fuzzy expert soft routing mechanism (fuzzy route), implemented via Fuzzy C-Means-based soft routing assignments, forming a new class of fuzzy MoE that leverages fuzzy expert clustering scores to dynamically integrate local (LocalSS) and global (GlobalSS) state space features. This approach enables effective balancing of global pixel dependencies and local texture information, thereby substantially reducing background interference and image noise in bright-field images and improving the accuracy of organoid identification. Furthermore, we incorporate a Dual Downsampling Adaptive Pooling Feature Fusion module, which combines original backbone features with parallel downsampled features and utilizes content-aware pooling for adaptive multilevel and multiscale feature fusion. Experimental results on multiclass organoid bright-field image datasets demonstrate that FEMSSorg achieves state-of-the-art performance in both organoid detection and morphological texture classification, highlighting its value as a robust computational tool for advancing real-time, high-throughput organoid research.

Drug-drug interaction (DDI) refers to the interaction relationships between drugs. Discovering new DDIs is crucial for advancing drug development and enhancing clinical treatments. Given the significant progress achieved through graph neural networks (GNNs), network-based models have become a prevalent approach for tackling this challenge. However, current network-based approaches are incapable of seamlessly integrating a wide range of information. Motivated by this discovery, we propose a novel model, namely LLM-DDI, which aims to comprehensively tackle DDI prediction tasks by integrating various information of molecules in the BKG. LLM-DDI initially incorporates the generative pre-trained transformer (GPT) model to generate embeddings for each molecule within the biomedical knowledge graph (BKG). These embeddings encompass diverse types of information pertaining to each molecule. Subsequently, LLM-DDI utilizes a message-passing GNN framework to enhance the learning of molecular representations with the embeddings derived from GPT as input. LLM-DDI governs the propagation of information within the BKG by semantic relationships. These semantic relationships determine how information flows and is exchanged between different entities in the BKG. Finally, LLM-DDI leverages the learned drug representations to predict potential DDIs. Experiments show the effectiveness of LLM-DDI, as it achieves the best performance on two real-world datasets, providing valuable guidance for drug development and clinical treatment.

Drug-drug interactions (DDIs) represent a critical challenge in pharmacology, often leading to adverse effects and compromised therapeutic efficacy. Accurate prediction of DDI events, which involve not only identifying interacting drug pairs but also characterizing the specific nature and context of their interactions, is essential for drug safety and personalized medicine. In this study, we propose a novel Multi-view Contrastive Learning framework, namely MCL-DDI, for DDI Event Prediction by leveraging multi-view representations of drugs to enhance predictive performance. MCL-DDI integrates molecular structures and network features, capturing complementary information about drug properties and interactions. By employing contrastive learning, we align and unify drug representations across these diverse views, enabling the framework to distinguish complex interaction patterns. Extensive experiments on benchmark datasets demonstrate that MCL-DDI outperforms state-of-the-art methods in terms of predictive accuracy. Furthermore, case studies highlight the model's ability to identify clinically relevant DDIs, offering practical insights for drug development and risk assessment. Our work establishes a robust and accurate paradigm for DDI event prediction, paving the way for safer and more effective pharmacological interventions.

Recent studies have demonstrated that miRNA expression dysregulation is closely related to the occurrence of various diseases; thus, miRNA-based drug development strategies have received increasing research interest. Most existing computational methods focus on the attribute information of individual nodes and are limited to the direct associations between nodes, thereby ignoring the complex associations inherent in the network. This limitation may lead to the loss of key potential information, which impacts the prediction accuracy. To address these issues, we propose a multisource information fusion and metapath enhancement matrix based graph autoencoder (MSMP-GAE) to predict the potential associations between miRNAs and drugs. The proposed MSMP-GAE model comprises a metapath instance extraction module, a metapath feature enhanced encoder module, a weighted feature fusion module, and a graph autoencoder. First, we construct an miRNA-drug heterogeneous network using experimentally validated miRNA-drug interactions and integrate various miRNA and drug features into an initial feature matrix to comprehensively represent their intrinsic property information. Then, we extract metapath instances from the interaction network, generate multiple metapath enhancement matrices, and fuse them with the initial feature matrix to generate high-quality node feature embeddings. Finally, we employ the graph autoencoder for fivefold cross-validation on a public dataset and test it on an independent test set. Experimental results demonstrate that the proposed MSMP-GAE model obtained an area under the curve (AUC) and AUPR values of 98.61% and 98.23%, respectively, which is considerably better than the several state-of-the-art methods. This highlights the importance of the higher-order complex associations between nodes in the miRNA-drug association (MDA) prediction task and provides a new method and approach to advance MDA prediction.

Accurately identifying cancer driver genes is crucial for understanding tumorigenesis and advancing precision oncology. However, integrating multi-omics data within complex biological networks remains challenging, particularly when it comes to capturing diverse structural information and leveraging the distinct signals from different omics modalities. While graphbased methods have demonstrated high accuracy in cancer gene identification, they might overlook the heterogeneity between omics features. To address this limitation, we propose CGI-MoE, a Mixture-of-Experts-inspired graph representation learning framework that incorporates omics-feature-specific expert modules based on graph transformers, together with adaptive gating and subgraph aggregation mechanisms. CGI-MoE extracts both local and global structural encodings for each node by sampling multiple subgraphs, enabling the model to capture comprehensive and robust network features. Each expert module focuses on a specific omics modality, and their outputs are fused by a lightweight gating network that dynamically weighs their contributions. When evaluated on both homogeneous and heterogeneous benchmark datasets, CGI-MoE achieves state-of-the-art performance in terms of accuracy, AUC, and AUPR, consistently surpassing existing methods. Ablation studies further highlight the critical roles of each component in achieving robust performance. Using the trained models, CGI-MoE predicted 46 novel cancer gene candidates from all unlabeled genes, demonstrating its potential for novel discovery and for deepening our understanding of cancer development. The code is available at https://github.com/moomight/CGI-MoE.

Traditional Chinese medicine (TCM), as a treasure of traditional Chinese medical heritage, has developed a unique therapeutic system through millennia of practice. Its multi-compound, multi-target mechanism demonstrates distinct advantages in disease treatment. However, under the modern medical framework, the modernization of TCM research faces dual challenges: on one hand, the complex interaction network between herbal compounds and target proteins has not yet been systematically elucidated, and traditional methods based on in vitro experiments or statistical correlations suffer from long experimental cycles and high failure rates; on the other hand, compared to chemical drugs with well-established large-scale annotated databases, the association data between herbal compounds and targets exhibit significant sparsity, making it difficult for existing methods to effectively model cross-domain knowledge transfer due to mismatched data dimensions. To address this scientific challenge, this study proposes a cross-domain association prediction framework, DGCPI, which integrates graph neural networks and adversarial learning. To begin with, DGCPI constructs two heterogeneous graph networks for modelling chemical drug-target (source domain) and herbal compound-target (taget domain) associations respectively. It then applies a graph convolutional network to capture source-domain-specific topological features in the chemical drug-target graph. After that, an adversarial discriminator is introduced to dynamically align the embedding distributions between the source and target domains, overcoming the dependence of traditional methods on the assumption of homomorphic data. Experimental results show that only with a small number of TCM samples involved in the training, the model significantly outperforms baseline models on two TCM datasets.

Traditional Chinese medicine (TCM), with its roots in centuries of clinical practice, has established itself as an effective therapeutic system that involves a diverse range of herbal plants. Despite its proven efficacy, the intricate relationships between herbal multi-component preparations and multi-target therapies present challenges for systematic study, thereby limiting its broader application in managing chronic diseases. In this work, we aim to identify novel therapeutic targets of natural compounds found in TCM herbs by leveraging advanced hypergraph representation learning techniques. Following the multi-component, multi-target pharmacological mechanisms, we first construct two hypergraphs to represent herb-compound and disease-target interactions, respectively. The connection between these hypergraphs is established through compound-target associations. A convolutional operator is then employed to capture the high-order correlations between compound (or target) nodes and herb (or disease) hyperedges within each hypergraph. Furthermore, we incorporate the PageRank algorithm and a multi-head attention mechanism to enhance the representation capabilities of node embeddings. By integrating these methods, our model is able to accurately identify novel therapeutic targets of natural compounds in TCM herbs in an end-to-end manner. Extensive experiments conducted on three benchmark datasets demonstrate the superior performance of our model when compared with several state-of-the-art approaches. Furthermore, case studies on two natural compounds, coumarin and progesterone, reveal that 7 and 8 out of the Top-10 identified targets, respectively, have been validated through literature review. These results highlight the effectiveness of our model in discovering new therapeutic targets for natural compounds in TCM.

Uncovering novel drug-drug interactions (DDIs) plays a pivotal role in advancing drug development and improving clinical treatment. The outstanding effectiveness of graph neural networks (GNNs) has garnered significant interest in the field of DDI prediction. Consequently, there has been a notable surge in the development of network-based computational approaches for predicting DDIs. However, current approaches face limitations in capturing the spatial relationships between neighboring nodes and their higher-level features during the aggregation of neighbor representations. To address this issue, this study introduces a novel model, KGCNN, designed to comprehensively tackle DDI prediction tasks by considering spatial relationships between molecules within the biomedical knowledge graph (BKG). KGCNN is built upon a message-passing GNN framework, consisting of propagation and aggregation. In the context of the BKG, KGCNN governs the propagation of information based on semantic relationships, which determine the flow and exchange of information between different molecules. In contrast to traditional linear aggregators, KGCNN introduces a spatial-aware capsule aggregator, which effectively captures the spatial relationships among neighboring molecules and their higher-level features within the graph structure. The ultimate goal is to leverage these learned drug representations to predict potential DDIs. To evaluate the effectiveness of KGCNN, it undergoes testing on two datasets. Extensive experimental results demonstrate its superiority in DDI predictions and quantified performance.

Advancements in high-throughput technologies have yielded large-scale human gut microbiota profiles, sparking considerable interest in exploring the relationship between the gut microbiome and complex human diseases. Through extracting and integrating knowledge from complex microbiome data, existing machine learning (ML)-based studies have demonstrated their effectiveness in the precise identification of high-risk individuals. However, these approaches struggle to address the heterogeneity and sparsity of microbial features and explore the intrinsic relatedness among human diseases. In this work, we reframe human gut microbiome-based disease detection as a multilabel classification (MLC) problem and integrate a range of innovative techniques within the proposed MLC framework, aptly named GutMLC. Specifically, the entity semantic similarity as priori knowledge is incorporated into multilabel feature selection and loss functions by capturing the shared attributes and inherent associations among diseases and microbes. To tackle the issue of label imbalance, both within and between labels, we adapt the focal loss (FL) function for MLC using debiased inverse weighting. Extensive experiment results consistently demonstrate the competitive performance of GutMLC in comparison with commonly used MLC and single-label classification (SLC) algorithms. This work seeks to unlock the potential of gut microbiota as robust biomarkers for multiple disease prediction.

To understand the mechanisms of complex systems, attributed graphs (AGs) are recognized as a valuable model by their capability of describing nontrivial topological structures and rich node contents, and their emergence raises new challenges on the task of graph clustering. Although a variety of computational algorithms have been proposed to perform accurate clustering analysis on AGs, most of them are incapable of inferring the cluster labels of nodes through links, thus falling short of explaining node behaviors on how to formulate overlapping clusters. Moreover, the vast amount of links considerably decreases the computation efficiency if they are explicitly taken into account for AG clustering. To overcome this problem, we present a novel variational Bayesian learning model, which avoids generating a complete AG by only simulating the generative process of its skeleton with the prior knowledge on the cluster labels of links. When addressing the inference problem, we develop an efficient algorithm, namely, LCAAG, for determining the optimal cluster labels of nodes by estimating local community structures of links. The convergence of LCAAG has been proved theoretically. Compared with several state-of-the-art algorithms, LCAAG has demonstrated its promising performance in terms of both accuracy and scalability on five different scaled benchmark datasets. The source code and datasets are available at https://github.com/shallowdreamoon/LCAAG.git.

Protein complexes play a crucial role in regulating various biological processes that govern cell activities. Numerous computational algorithms have been proposed to identify protein complexes from protein-protein interaction (PPI) networks. However, many of these algorithms face limitations in effectively leveraging multiview biological information of proteins, restricting their ability to capture the intricate characteristics of protein complexes in PPI networks. While deep learning-based algorithms have significantly advanced the identification of protein complexes, they often integrate graph representation learning techniques into traditional clustering algorithms without explicitly capturing the dependency between protein embeddings and resulting complexes. To address these issues, we present a multiview fusion neural network, named FMvPCI, for protein complex identification via fuzzy clustering. In FMvPCI, we introduce a novel multiview graph convolution encoder to effectively manipulate and fuse the biological information of proteins from different perspectives. Subsequently, the optimization of FMvPCI incorporates our expectations about protein complexes through the concept of fuzzy clustering. This approach unifies the embeddings of proteins and their cluster memberships within a coherent framework. Leveraging a heuristic search strategy, FMvPCI can discover overlapping protein complexes based on the cluster memberships of proteins. A series of experiments on five different PPI networks collected from two species have been conducted to evaluate the performance of FMvPCI by comparing it with state-of-the-art identification algorithms, and the results demonstrate the superior performance of FMvPCI by significantly improving the identification accuracy for protein complexes.

Identifying novel drug-drug interactions (DDIs) is a crucial task in pharmacology, as the interference between pharmacological substances can pose serious medical risks. In recent years, several network-based techniques have emerged for predicting DDIs. However, they primarily focus on local structures within DDI-related networks, often overlooking the significance of indirect connections between pairwise drug nodes from a global perspective. Additionally, effectively handling heterogeneous information present in both biomedical knowledge graphs and drug molecular graphs remains a challenge for improved performance of DDI prediction. To address these limitations, we propose a Transformer-based relatIon-aware Graph rEpresentation leaRning framework (TIGER) for DDI prediction. TIGER leverages the Transformer architecture to effectively exploit the structure of heterogeneous graph, which allows it direct learning of long dependencies and high-order structures. Furthermore, TIGER incorporates a relation-aware self-attention mechanism, capturing a diverse range of semantic relations that exist between pairs of nodes in heterogeneous graph. In addition to these advancements, TIGER enhances predictive accuracy by modeling DDI prediction task using a dual-channel network, where drug molecular graph and biomedical knowledge graph are fed into two respective channels. By incorporating embeddings obtained at graph and node levels, TIGER can benefit from structural properties of drugs as well as rich contextual information provided by biomedical knowledge graph. Extensive experiments conducted on three real-world datasets demonstrate the effectiveness of TIGER in DDI prediction. Furthermore, case studies highlight its ability to provide a deeper understanding of underlying mechanisms of DDIs.

Organoids, 3D in vitro cultures replicating their tissue or organ of origin, hold great potential for disease research, drug testing, and regenerative medicine. However, current analysis methods primarily rely on staining techniques, which limit high-throughput analysis and require manual adjustments. Automated and non-invasive methods remain underexplored, and most rely on small-scale datasets that fail to capture the complexity of organoids, which vary significantly in shape and size and have more intricate backgrounds than cells. To address these limitations, we developed MSSOrg, a multiscale framework for organoid localization and classification, integrating Local and Global State Space Models (SSMs) to balance pixel dependencies and texture features. MSSOrg mitigates background interference and image noise in bright-field images, achieving superior performance in both organoid detection and morphological classification. This innovation enables real-time, large-scale organoid analysis, providing a powerful tool for drug screening and disease research, ultimately accelerating progress in these fields.

Exploring drug-drug interactions (DDIs) is crucial for avoiding unknown physicochemical incompatibilities between coadministered drugs. While most studies concentrate on detecting the presence or absence of DDIs, they often overlook the diversity of DDI event types that can significantly enhance drug research and guide scientific drug use. To address this limitation, we propose MNCLDDI, a multi-view nested contrastive learning model designed for the precise prediction of DDI events. MN-CLDDI begins by employing a relational graph convolutional network to capture the various explicit relationships between drugs within a multi-relational DDI graph. This is followed by a transformer framework combined with a convolutional neural network (CNN) to learn the biological features of drugs from their Smiles information. The model then integrates these two feature types into a novel multi-view nested contrastive learning framework, thereby improving the expressiveness of drug embeddings from multiple biological perspectives. Experimental results on two real-world datasets demonstrate that MNCLDDI outperforms state-of-the-art models in predicting DDI events. Moreover, our case studies reveal that considering the multi-view features of drugs simultaneously enables MNCLDDI to predict DDI events with greater accuracy and from a more comprehensive perspective, offering valuable insights into the study of DDI events.

Numerous studies have demonstrated that the interaction between miRNAs and drugs plays a pivotal role in regulating gene expression and cellular function. Therefore, predicting these interactions is crucial for the development of novel drugs and personalized therapies. Existing methods for predicting miRNA-drug interactions often fail to leverage the full spectrum of molecular and biological features and overlook complex high-dimensional patterns. Deep non-negative matrix factorization (DNMF) addresses these limitations by extracting higher-level representations, thereby enhancing prediction accuracy and robustness. Building on this, this paper proposes a model called DNMDA. In this model, we integrate multiple similarity networks for both miRNAs and drugs and then extract their features through three key modules. Moreover, autoencoders are used to combine various feature sets, allowing for the capture of complementary information and enhancing the model’s capacity for making precise and reliable predictions. The resulting features are consolidated into a unified feature vector for each miRNA-drug pair. Ultimately, these feature vectors and their associated labels are provided to the classifier for training. To verify the predictions, a five-fold cross-validation was conducted. The five-fold cross-validation demonstrated a clear advantage in DNMDA’s metrics, underscoring its reliability in predicting potential miRNA-drug interactions. This claim is further supported by the predictive results section in the paper, offering concrete evidence of DNMDA’s efficacy in this field.

Traditional Chinese medicine (TCM) has a long history of herbal treatments for diseases, but our understanding to the interactions between herbal compounds and target proteins remains considerably incomplete due to the complexity of multi-compound, multi-target mechanisms. Most of existing prediction models fall short of considering such mechanisms, limiting their applicability in discovering novel compound-target interactions(CTIs). To address this problem, we propose DHGCTI, a novel Hypergraph Convolutional Network for improving performance on the task of CTI prediction. In the context of hypergraph, herbs and diseases are regarded as the hyperedges of herbal compounds and target proteins respectively. An end-to-end prediction model is then specifically designed to identify CTIs by using a hypergraph convolutional network. Extensive testing shows experimental results demonstrate that DHGCTI consistently outperforms baseline models, highlighting its superior performance. Moreover, we conducted case studies on acacetin and luteolin, and 6 and 7 of the top ten scoring targets were validated by literature, respectively. This highlights the advantages of our model in exploring new targets of natural compounds in traditional Chinese medicine.

Protein complexes are essential in regulating various cellular processes. A number of computational algorithms have been developed to identify protein complexes from protein-protein interaction (PPI) networks, but they are limited in their ability to effectively leverage diverse biological knowledge of proteins. Additionally, while deep learning-based algorithms perform well in identifying protein complexes, they fail to explicitly capture the dependency between protein embeddings and resulting complexes. To address these challenges, this paper proposes a knowledge-guided protein complex identification algorithm with fuzzy-based graph representation learning, named KPCI-FGRL. In particular, a fuzzy-based graph representation learning framework is developed by KPCI-FGRL to manipulate and fuse network structure with multi-view biological knowledge of proteins. During the training phase of KPCI-FGRL, besides employing self-supervised loss to improve the cohesion of the complexes, we also specifically incorporate the expectation about protein complexes based on fuzzy clustering concept, and thus the dependency between protein embeddings and complexes can be coupled. Furthermore, KPCI-FGRL is capable of achieving the identification of overlapping protein complexes through a heuristic search strategy upon fuzzy memberships of proteins. Extensive experimental results on four different PPI networks collected from two species demonstrate that KPCI-FGRL significantly outperforms several state-of-the-art protein complex identification algorithms.

Herbs applicability in disease treatment has been verified through experiences over thousands of years. The understanding of herb-disease associations (HDAs) is yet far from complete due to the complicated mechanism inherent in multi-target and multi-component (MTMC) botanical therapeutics. Most of the existing prediction models fail to incorporate the MTMC mechanism. To overcome this problem, we propose a novel dual-channel hypergraph convolutional network, namely HGHDA, for HDA prediction. Technically, HGHDA first adopts an autoencoder to project components and target protein onto a low-dimensional latent space so as to obtain their embeddings by preserving similarity characteristics in their original feature spaces. To model the high-order relations between herbs and their components, we design a channel in HGHDA to encode a hypergraph that describes the high-order patterns of herb-component relations via hypergraph convolution. The other channel in HGHDA is also established in the same way to model the high-order relations between diseases and target proteins. The embeddings of drugs and diseases are then aggregated through our dual-channel network to obtain the prediction results with a scoring function. To evaluate the performance of HGHDA, a series of extensive experiments have been conducted on two benchmark datasets, and the results demonstrate the superiority of HGHDA over the state-of-the-art algorithms proposed for HDA prediction. Besides, our case study on Chuan Xiong and Astragalus membranaceus is a strong indicator to verify the effectiveness of HGHDA, as seven and eight out of the top 10 diseases predicted by HGHDA for Chuan-Xiong and Astragalus-membranaceus, respectively, have been reported in literature.

Attributed graph (AG) clustering is a fundamental, yet challenging, task for studying underlying network structures. Recently, a variety of graph representation learning models has been proposed to effectively infer the node embeddings, which are then incorporated into conventional clustering techniques to identify meaningful clusters. While these models tend to preserve node proximities, which reflect the similarity between nodes in both structural and attribute dimensions, for representation learning, they generally overlook the crucial dependencies between node embeddings and the resulting clusters. To overcome this problem, we propose a novel fuzzy-based deep AG clustering model, namely FDAGC, which is capable of achieving the task in a purely unsupervised and end-to-end manner without additionally incorporating conventional clustering techniques. In particular, FDAGC first encodes network structures and node attributes into a compact representation with graph convolution. A reconstruction error is then estimated to minimize the information loss during network message-passing. Besides, we utilize a self-monitoring training strategy to optimize node embeddings, thus improving the cluster cohesion by guiding them toward cluster centers. In the training phase, our expectations about resulting clusters are explicitly incorporated into the optimization of FDAGC via the concept of fuzzy clustering, thus leading to more accurate clustering by coupling the dependency between graph representation learning and AG clustering. Extensive experiments have demonstrated the superior performance of FDAGC in terms of several evaluation metrics, such as accuracy, normalized mutual information, F1-score and adjusted rand index, on six real-world AGs with different scales.

In the field of artificial intelligence for information technology operations, operational data are often modeled as aperiodic multivariate time series, which contain rich multidimensional and nonlinear patterns. However, the existing approaches are unable to effectively acquire knowledge and recognize patterns due to their reliance on processing and modeling periodic patterns. To address this issue, this article proposes a novel deep-saliency-aware dual embedded attention network for aperiodic multivariate time-series forecasting. Our network consists of three main components: 1) a convolutional-neural-network- and transformer-based component for saliency representation of the aperiodic patterns; 2) a lightweight recurrent neural network component for capturing long-term dependence features; and 3) an attention mechanism for fusing the latent representations from the former components. Extensive empirical studies are conducted on a real-world dataset and five other public datasets to evaluate the proposed network against four state-of-the-art models. The results show that our method achieves impressive high performance on most evaluation metrics. Furthermore, the data and code used in this study are publicly available, which can facilitate progress in the community.

As a pivotal post-transcriptional modification of RNA, N6-methyladenosine (m6A) has a substantial influence on gene expression modulation and cellular fate determination. Although a variety of computational models have been developed to accurately identify potential m6A modification sites, few of them are capable of interpreting the identification process with insights gained from consensus knowledge. To overcome this problem, we propose a deep learning model, namely M6A-DCR, by discovering consensus regions for interpretable identification of m6A modification sites. In particular, M6A-DCR first constructs an instance graph for each RNA sequence by integrating specific positions and types of nucleotides. The discovery of consensus regions is then formulated as a graph clustering problem in light of aggregating all instance graphs. After that, M6A-DCR adopts a motif-aware graph reconstruction optimization process to learn high-quality embeddings of input RNA sequences, thus achieving the identification of m6A modification sites in an end-to-end manner. Experimental results demonstrate the superior performance of M6A-DCR by comparing it with several state-of-the-art identification models. The consideration of consensus regions empowers our model to make interpretable predictions at the motif level. The analysis of cross validation through different species and tissues further verifies the consistency between the identification results of M6A-DCR and the evolutionary relationships among species.

As post-transcriptional regulators of gene expression, micro-ribonucleic acids (miRNAs) are regarded as potential biomarkers for a variety of diseases. Hence, the prediction of miRNA-disease associations (MDAs) is of great significance for an in-depth understanding of disease pathogenesis and progression. Existing prediction models are mainly concentrated on incorporating different sources of biological information to perform the MDA prediction task while failing to consider the fully potential utility of MDA network information at the motif-level. To overcome this problem, we propose a novel motif-aware MDA prediction model, namely MotifMDA, by fusing a variety of high- and low-order structural information. In particular, we first design several motifs of interest considering their ability to characterize how miRNAs are associated with diseases through different network structural patterns. Then, MotifMDA adopts a two-layer hierarchical attention to identify novel MDAs. Specifically, the first attention layer learns high-order motif preferences based on their occurrences in the given MDA network, while the second one learns the final embeddings of miRNAs and diseases through coupling high- and low-order preferences. Experimental results on two benchmark datasets have demonstrated the superior performance of MotifMDA over several state-of-the-art prediction models. This strongly indicates that accurate MDA prediction can be achieved by relying solely on MDA network information. Furthermore, our case studies indicate that the incorporation of motif-level structure information allows MotifMDA to discover novel MDAs from different perspectives.

In medical research, organoids, which exhibit structural and functional resemblances to authentic organs, offer a substantial avenue for delving into aspects encompassing physiology, pathophysiology, diseases, and pharmaceutical screening. Critical insights into drug responsiveness are often gleaned from these organoids' dimensional and configurational disparities. However, conventional detection methodologies reliant upon fluorescent labeling engender potential hazards to organoid integrity, thereby impinging upon their intrinsic dynamic attributes. Traditional bounding-box detection methodologies fall short in encapsulating intricate morphological particulars, and certain deep-learning approaches grapple with the intricate task of capturing multi-scale data, particularly when tasked with discerning organoid structures characterized by marked shape and size heterogeneities. In a bid to surmount these constraints, our study introduces CAMPEOD, an innovative framework that synergistically amalgamates multi-scale attributes derived from organoid specimens, employing cross-attention mechanisms. This novel approach effectively obviates superfluous background interference and image noise, thereby endowing an automated, finely-tuned dissection of organoid samples. Significantly, this segmentation process ensures congruence with authentic organoid quantities and morphological characteristics. By facilitating comprehensive scrutiny of microscopy images of organoid samples on a large scale, CAMPEOD assumes considerable implications for the realm of pharmaceutical screening and ailment emulation.

The utilization of web social media for depression detection has been proven effective in recent years since the multimedia signal on web can reflect users’ emotions, feelings, and personality traits in advance. However, most earlier studies simply used users’ submitted words or user profiles to predict depression risk. The implicit information accessible in users’ posted images, which can be effective in depression detection, still remains unexplored. In this paper, an implicit and explicit multi-modal feature fusion (IEMFF) model is proposed for depression detection. We successfully make the implicit information inherent in users’ posted images explicit and further incorporate such explicit features with the textual features directly extracted from user-posted texts. A multi-modal feature fusion approach is applied for depression detection. Extensive experiments have been conducted on public Twitter datasets. Experimental results show that our approach has achieved state-of-the-art performance for depression detection.

The approaches based on deep learning have achieved remarkable success in diabetic retinopathy detection. Due to the accountability in medical diagnosis, the interpretability of computer-aided diagnosis has recently been investigated. However, few existing approaches make full use of the explainable evidence to improve the diagnosis accuracy. In this paper, we propose a Model-free Lesion Generation and Learning (MLGL) framework to study the interpretability of diabetic retinopathy detection. We first generate visual explanations for diabetic retinopathy diagnosis using the proposed Gated Multi-layer Saliency Map (GMSM) module, which locates the accurate region of lesions by combining multi-layer heatmaps. Then we use the GMSM to extract the lesion patches and conduct the adaptive lesion transfer, iteratively generating new retinal fundus images with lesions. Especially, in this process, no additional generative models are trained. Finally, we merge the generated and original retinal fundus images for the model's training to learn robust lesion features. Overall, our method provides accurate explainable evidence and further addresses the data imbalance problem in diabetic retinopathy detection. The experimental results on four public datasets demonstrate the efficiency of our approach.

N6-methyladenosine (m6A) regulates RNA post-transcriptional modification and translation processes, thereby regulating gene expression and cell fate. Hence, accurate identification of potential m6A modification sites is a key step to further reveal their biological functions and understand multiple biological processes such as gene regulation and epigenetic variation. Many computational methods have been developed to address this challenge. However, fewer studies have focused on an interpretable process of m6A modification site identification. Here, we propose an interpretable end-to-end predictor, called M6AInter, which learns the RNA sequence patterns related to modification sites through contrastive learning frameworks to achieve accurate identification of m6A modification sites. Specifically, M6AInter first utilizes chaos game representation theory and one-hot encoding to initialize the position and type information of nucleotides, respectively. On this basis, M6AInter extracts the position and type correlations shared by RNA sequences, and predicts the common sequence patterns by utilizing a graph contrastive clustering framework. These motifs and patterns are involved in describing the associations between RNA sequences and obtaining their low-dimensional representations. Finally, through a designed bias fusion block, these representations are combined with the frequency information of nucleotides to realize the identification of m6A modification sites. Extensive experimental results show that our model can accurately identify modified RNA sequences and can adaptively locate sequential regions associated with m6A modification sites on RNA sequences. Importantly, by exploring the role of these patterns in the identification tasks, M6AInter provides interpretable predictions and analysis at the sequence level.

MOTIVATION: The task of predicting drug-target interactions (DTIs) plays a significant role in facilitating the development of novel drug discovery. Compared with laboratory-based approaches, computational methods proposed for DTI prediction are preferred due to their high-efficiency and low-cost advantages. Recently, much attention has been attracted to apply different graph neural network (GNN) models to discover underlying DTIs from heterogeneous biological information network (HBIN). Although GNN-based prediction methods achieve better performance, they are prone to encounter the over-smoothing simulation when learning the latent representations of drugs and targets with their rich neighborhood information in HBIN, and thereby reduce the discriminative ability in DTI prediction. RESULTS: In this work, an improved graph representation learning method, namely iGRLDTI, is proposed to address the above issue by better capturing more discriminative representations of drugs and targets in a latent feature space. Specifically, iGRLDTI first constructs an HBIN by integrating the biological knowledge of drugs and targets with their interactions. After that, it adopts a node-dependent local smoothing strategy to adaptively decide the propagation depth of each biomolecule in HBIN, thus significantly alleviating over-smoothing by enhancing the discriminative ability of feature representations of drugs and targets. Finally, a Gradient Boosting Decision Tree classifier is used by iGRLDTI to predict novel DTIs. Experimental results demonstrate that iGRLDTI yields better performance that several state-of-the-art computational methods on the benchmark dataset. Besides, our case study indicates that iGRLDTI can successfully identify novel DTIs with more distinguishable features of drugs and targets. AVAILABILITY AND IMPLEMENTATION: Python codes and dataset are available at https://github.com/stevejobws/iGRLDTI/.

BACKGROUND: As an important task in bioinformatics, clustering analysis plays a critical role in understanding the functional mechanisms of many complex biological systems, which can be modeled as biological networks. The purpose of clustering analysis in biological networks is to identify functional modules of interest, but there is a lack of online clustering tools that visualize biological networks and provide in-depth biological analysis for discovered clusters. RESULTS: Here we present BioCAIV, a novel webserver dedicated to maximize its accessibility and applicability on the clustering analysis of biological networks. This, together with its user-friendly interface, assists biological researchers to perform an accurate clustering analysis for biological networks and identify functionally significant modules for further assessment. CONCLUSIONS: BioCAIV is an efficient clustering analysis webserver designed for a variety of biological networks. BioCAIV is freely available without registration requirements at http://bioinformatics.tianshanzw.cn:8888/BioCAIV/ .

Interactions between transcription factor and target gene form the main part of gene regulation network in human, which are still complicating factors in biological research. Specifically, for nearly half of those interactions recorded in established database, their interaction types are yet to be confirmed. Although several computational methods exist to predict gene interactions and their type, there is still no method available to predict them solely based on topology information. To this end, we proposed here a graph-based prediction model called KGE-TGI and trained in a multi-task learning manner on a knowledge graph that we specially constructed for this problem. The KGE-TGI model relies on topology information rather than being driven by gene expression data. In this paper, we formulate the task of predicting interaction types of transcript factor and target genes as a multi-label classification problem for link types on a heterogeneous graph, coupled with solving another link prediction problem that is inherently related. We constructed a ground truth dataset as benchmark and evaluated the proposed method on it. As a result of the 5-fold cross experiments, the proposed method achieved average AUC values of 0.9654 and 0.9339 in the tasks of link prediction and link type classification, respectively. In addition, the results of a series of comparison experiments also prove that the introduction of knowledge information significantly benefits to the prediction and that our methodology achieve state-of-the-art performance in this problem.

Protein-protein interactions (PPIs) play a critical role in the proteomics study, and a variety of computational algorithms have been developed to predict PPIs. Though effective, their performance is constrained by high false-positive and false-negative rates observed in PPI data. To overcome this problem, a novel PPI prediction algorithm, namely PASNVGA, is proposed in this work by combining the sequence and network information of proteins via variational graph autoencoder. To do so, PASNVGA first applies different strategies to extract the features of proteins from their sequence and network information, and obtains a more compact form of these features using principal component analysis. In addition, PASNVGA designs a scoring function to measure the higher-order connectivity between proteins and so as to obtain a higher-order adjacency matrix. With all these features and adjacency matrices, PASNVGA trains a variational graph autoencoder model to further learn the integrated embeddings of proteins. The prediction task is then completed by using a simple feedforward neural network. Extensive experiments have been conducted on five PPI datasets collected from different species. Compared with several state-of-the-art algorithms, PASNVGA has been demonstrated as a promising PPI prediction algorithm.

Identifying protein targets for drugs establishes an indispensable knowledge foundation for drug repurposing and drug development. Though expensive and time-consuming, vitro trials are widely employed to discover drug targets, and the existing relevant computational algorithms still cannot satisfy the demand for real application in drug R&D with regards to the prediction accuracy and performance efficiency, which are urgently needed to be improved. To this end, we propose here the PPAEDTI model, which uses the graph personalized propagation technique to predict drug-target interactions from the known interaction network. To evaluate the prediction performance, six benchmark datasets were used for testing with some state-of-the-art methods compared. As a result, using the 5-fold cross-validation, the proposed PPAEDTI model achieves average AUCs>90% on 5 collected datasets. We also manually checked the top-20 prediction list for 2 proteins (hsa:775 and hsa:779) and a kind of drug (D00618), and successfully confirmed 18, 17, and 20 items from the public datasets, respectively. The experimental results indicate that, given known drug-target interactions, the PPAEDTI model can provide accurate predictions for the new ones, which is anticipated to serve as a useful tool for pharmacology research. Using the proposed model that was trained with the collected datasets, we have built a computational platform that is accessible at http://120.77.11.78/PPAEDTI/ and corresponding codes and datasets are also released.

Identifying Drug-Target Interactions (DTIs) is a critical step in studying pathogenesis and drug development. Due to the fact that conventional experimental methods usually suffer from high costs and low efficiency, various computational methods have been proposed to detect potential DTIs by extracting features from the biological information of drugs and their target proteins. Though effective, most of them fall short of considering the topological structure of the DTI network, which provides a global view to discover novel DTIs. In this paper, a network-based computational method, namely LG-DTI, is proposed to accurately predict DTIs over a heterogeneous information network. For drugs and target proteins, LG-DTI first learns not only their local representations from drug molecular structures and protein sequences, but also their global representations by using a semi-supervised heterogeneous network embedding method. These two kinds of representations consist of the final representations of drugs and target proteins, which are then incorporated into a Random Forest classifier to complete the task of DTI prediction. The performance of LG-DTI has been evaluated on two independent datasets and also compared with several state-of-the-art methods. Experimental results show the superior performance of LG-DTI. Moreover, our case study indicates that LG-DTI can be a valuable tool for identifying novel DTIs.

The outbreak of COVID-19 caused by SARS-coronavirus (CoV)-2 has made millions of deaths since 2019. Although a variety of computational methods have been proposed to repurpose drugs for treating SARS-CoV-2 infections, it is still a challenging task for new viruses, as there are no verified virus-drug associations (VDAs) between them and existing drugs. To efficiently solve the cold-start problem posed by new viruses, a novel constrained multi-view nonnegative matrix factorization (CMNMF) model is designed by jointly utilizing multiple sources of biological information. With the CMNMF model, the similarities of drugs and viruses can be preserved from their own perspectives when they are projected onto a unified latent feature space. Based on the CMNMF model, we propose a deep learning method, namely VDA-DLCMNMF, for repurposing drugs against new viruses. VDA-DLCMNMF first initializes the node representations of drugs and viruses with their corresponding latent feature vectors to avoid a random initialization and then applies graph convolutional network to optimize their representations. Given an arbitrary drug, its probability of being associated with a new virus is computed according to their representations. To evaluate the performance of VDA-DLCMNMF, we have conducted a series of experiments on three VDA datasets created for SARS-CoV-2. Experimental results demonstrate that the promising prediction accuracy of VDA-DLCMNMF. Moreover, incorporating the CMNMF model into deep learning gains new insight into the drug repurposing for SARS-CoV-2, as the results of molecular docking experiments reveal that four antiviral drugs identified by VDA-DLCMNMF have the potential ability to treat SARS-CoV-2 infections.

Drug-drug interactions (DDIs) are known as the main cause of life-threatening adverse events, and their identification is a key task in drug development. Existing computational algorithms mainly solve this problem by using advanced representation learning techniques. Though effective, few of them are capable of performing their tasks on biomedical knowledge graphs (KGs) that provide more detailed information about drug attributes and drug-related triple facts. In this work, an attention-based KG representation learning framework, namely DDKG, is proposed to fully utilize the information of KGs for improved performance of DDI prediction. In particular, DDKG first initializes the representations of drugs with their embeddings derived from drug attributes with an encoder-decoder layer, and then learns the representations of drugs by recursively propagating and aggregating first-order neighboring information along top-ranked network paths determined by neighboring node embeddings and triple facts. Last, DDKG estimates the probability of being interacting for pairwise drugs with their representations in an end-to-end manner. To evaluate the effectiveness of DDKG, extensive experiments have been conducted on two practical datasets with different sizes, and the results demonstrate that DDKG is superior to state-of-the-art algorithms on the DDI prediction task in terms of different evaluation metrics across all datasets.

Drug repositioning (DR) is a promising strategy to discover new indicators of approved drugs with artificial intelligence techniques, thus improving traditional drug discovery and development. However, most of DR computational methods fall short of taking into account the non-Euclidean nature of biomedical network data. To overcome this problem, a deep learning framework, namely DDAGDL, is proposed to predict drug-drug associations (DDAs) by using geometric deep learning (GDL) over heterogeneous information network (HIN). Incorporating complex biological information into the topological structure of HIN, DDAGDL effectively learns the smoothed representations of drugs and diseases with an attention mechanism. Experiment results demonstrate the superior performance of DDAGDL on three real-world datasets under 10-fold cross-validation when compared with state-of-the-art DR methods in terms of several evaluation metrics. Our case studies and molecular docking experiments indicate that DDAGDL is a promising DR tool that gains new insights into exploiting the geometric prior knowledge for improved efficacy.

BACKGROUND: The site information of substrates that can be cleaved by human immunodeficiency virus 1 proteases (HIV-1 PRs) is of great significance for designing effective inhibitors against HIV-1 viruses. A variety of machine learning-based algorithms have been developed to predict HIV-1 PR cleavage sites by extracting relevant features from substrate sequences. However, only relying on the sequence information is not sufficient to ensure a promising performance due to the uncertainty in the way of separating the datasets used for training and testing. Moreover, the existence of noisy data, i.e., false positive and false negative cleavage sites, could negatively influence the accuracy performance. RESULTS: In this work, an ensemble learning algorithm for predicting HIV-1 PR cleavage sites, namely EM-HIV, is proposed by training a set of weak learners, i.e., biased support vector machine classifiers, with the asymmetric bagging strategy. By doing so, the impact of data imbalance and noisy data can thus be alleviated. Besides, in order to make full use of substrate sequences, the features used by EM-HIV are collected from three different coding schemes, including amino acid identities, chemical properties and variable-length coevolutionary patterns, for the purpose of constructing more relevant feature vectors of octamers. Experiment results on three independent benchmark datasets demonstrate that EM-HIV outperforms state-of-the-art prediction algorithm in terms of several evaluation metrics. Hence, EM-HIV can be regarded as a useful tool to accurately predict HIV-1 PR cleavage sites.

BACKGROUND: Protein-protein interaction (PPI) plays an important role in regulating cells and signals. Despite the ongoing efforts of the bioassay group, continued incomplete data limits our ability to understand the molecular roots of human disease. Therefore, it is urgent to develop a computational method to predict PPIs from the perspective of molecular system. METHODS: In this paper, a highly efficient computational model, MTV-PPI, is proposed for PPI prediction based on a heterogeneous molecular network by learning inter-view protein sequences and intra-view interactions between molecules simultaneously. On the one hand, the inter-view feature is extracted from the protein sequence by k-mer method. On the other hand, we use a popular embedding method LINE to encode the heterogeneous molecular network to obtain the intra-view feature. Thus, the protein representation used in MTV-PPI is constructed by the aggregation of its inter-view feature and intra-view feature. Finally, random forest is integrated to predict potential PPIs. RESULTS: To prove the effectiveness of MTV-PPI, we conduct extensive experiments on a collected heterogeneous molecular network with the accuracy of 86.55%, sensitivity of 82.49%, precision of 89.79%, AUC of 0.9301 and AUPR of 0.9308. Further comparison experiments are performed with various protein representations and classifiers to indicate the effectiveness of MTV-PPI in predicting PPIs based on a complex network. CONCLUSION: The achieved experimental results illustrate that MTV-PPI is a promising tool for PPI prediction, which may provide a new perspective for the future interactions prediction researches based on heterogeneous molecular network.

Protein complexes are of great significance to provide valuable insights into the mechanisms of biological processes of proteins. A variety of computational algorithms have thus been proposed to identify protein complexes in a protein-protein interaction network. However, few of them can perform their tasks by taking into account both network topology and protein attribute information in a unified fuzzy-based clustering framework. Since proteins in the same complex are similar in terms of their attribute information and the consideration of fuzzy clustering can also make it possible for us to identify overlapping complexes, we target to propose such a novel fuzzy-based clustering framework, namely FCAN-PCI, for an improved identification accuracy. To do so, the semantic similarity between the attribute information of proteins is calculated and we then integrate it into a well-established fuzzy clustering model together with the network topology. After that, a momentum method is adopted to accelerate the clustering procedure. FCAN-PCI finally applies a heuristical search strategy to identify overlapping protein complexes. A series of extensive experiments have been conducted to evaluate the performance of FCAN-PCI by comparing it with state-of-the-art identification algorithms and the results demonstrate the promising performance of FCAN-PCI.

Proteins interact with each other to play critical roles in many biological processes in cells. Although promising, laboratory experiments usually suffer from the disadvantages of being time-consuming and labor-intensive. The results obtained are often not robust and considerably uncertain. Due recently to advances in high-throughput technologies, a large amount of proteomics data has been collected and this presents a significant opportunity and also a challenge to develop computational models to predict protein-protein interactions (PPIs) based on these data. In this paper, we present a comprehensive survey of the recent efforts that have been made towards the development of effective computational models for PPI prediction. The survey introduces the algorithms that can be used to learn computational models for predicting PPIs, and it classifies these models into different categories. To understand their relative merits, the paper discusses different validation schemes and metrics to evaluate the prediction performance. Biological databases that are commonly used in different experiments for performance comparison are also described and their use in a series of extensive experiments to compare different prediction models are discussed. Finally, we present some open issues in PPI prediction for future work. We explain how the performance of PPI prediction can be improved if these issues are effectively tackled.

MOTIVATION: Identifying microRNAs that are associated with different diseases as biomarkers is a problem of great medical significance. Existing computational methods for uncovering such microRNA-diseases associations (MDAs) are mostly developed under the assumption that similar microRNAs tend to associate with similar diseases. Since such an assumption is not always valid, these methods may not always be applicable to all kinds of MDAs. Considering that the relationship between long noncoding RNA (lncRNA) and different diseases and the co-regulation relationships between the biological functions of lncRNA and microRNA have been established, we propose here a multiview multitask method to make use of the known lncRNA-microRNA interaction to predict MDAs on a large scale. The investigation is performed in the absence of complete information of microRNAs and any similarity measurement for it and to the best knowledge, the work represents the first ever attempt to discover MDAs based on lncRNA-microRNA interactions. RESULTS: In this paper, we propose to develop a deep learning model called MVMTMDA that can create a multiview representation of microRNAs. The model is trained based on an end-to-end multitasking approach to machine learning so that, based on it, missing data in the side information can be determined automatically. Experimental results show that the proposed model yields an average area under ROC curve of 0.8410+/-0.018, 0.8512+/-0.012 and 0.8521+/-0.008 when k is set to 2, 5 and 10, respectively. In addition, we also propose here a statistical approach to predicting lncRNA-disease associations based on these associations and the MDA discovered using MVMTMDA. AVAILABILITY: Python code and the datasets used in our studies are made available at https://github.com/yahuang1991polyu/MVMTMDA/.

Network cluster analysis is of great importance as it is closely related to diverse applications, such as social community detection, biological module identification, and document segmentation. Aiming to effectively uncover clusters in the network data, a number of computational approaches , which utilize network topology , single vector of vertex features, or both the aforementioned, have been proposed. However, most prevalent approaches are incapable of dealing with those contemporary network data whose vertices are characterized by features collected from multiple sources. To address this challenge, in this paper, we propose a novel framework, dubbed Multi-Source Propagation Aware Network Clustering (MSPANC) for uncovering clusters in network data possessing multiple sources of vertex features. Different from most previous approaches, MSPANC is able to infer the cluster preference for each vertex utilizing both network topology and multi-source vertex features. To improve the practical significance of the discovered clusters, the learning of cluster membership is also involved into the modeling of the maximization of intra-cluster propagation regarding multi-source features. We propose a unified objective function for MSPANC to perform the clustering task and derive an alternative manner of learning algorithm for model optimization. Besides, we theoretically prove the convergence of the algorithm for optimizing MSPANC. The proposed model has been tested on five real-world datasets, including social, biological and document networks, and has been compared with several competitive baselines. The remarkable experimental results validate the effectiveness of MSPANC.

Neural network-based sequence-to-sequence (seq2seq) models strongly suffer from the lowdiversity problem when it comes to opendomain dialogue generation.As bland and generic utterances usually dominate the frequency distribution in our daily chitchat, avoiding them to generate more interesting responses requires complex data filtering, sampling techniques or modifying the training objective.In this paper, we propose a new perspective to diversify dialogue generation by leveraging non-conversational text.Compared with bilateral conversations, nonconversational text are easier to obtain, more diverse and cover a much broader range of topics.We collect a large-scale nonconversational corpus from multi sources including forum comments, idioms and book snippets.We further present a training paradigm to effectively incorporate these text via iterative back translation.The resulting model is tested on two conversational datasets and is shown to produce significantly more diverse responses without sacrificing the relevance with context.

The use of social media runs through our lives, and users' emotions are also affected by it. Previous studies have reported social organizations and psychologists using social media to find depressed patients. However, due to the variety of content published by users, it isn't effortless for the system to consider the text, image, and even the hidden information behind the image. To address this problem, we proposed a new system for social media screening of depressed patients named BlueMemo. We collected real-time posts from Twitter. Based on the posts, learned text features, image features, and visual attributes were extracted as three modalities and were fed into a multi-modal fusion and classification model to implement our system. The proposed BlueMemo has the power to help physicians and clinicians quickly and accurately identify users at potential risk for depression.

More than 300 million people have been affected by depression all over the world. Due to the medical equipment and knowledge limitations, most of them are not diagnosed at the early stages. Recent work attempts to use social media to detect depression since the patterns of opinions and thoughts expression of the posted text and images, can reflect users' mental state to some extent. In this work, we design a system dubbed SenseMood to demonstrate that the users with depression can be efficiently detected and analyzed by using proposed system. A deep visual-textual multimodal learning approach has been proposed to reveal the psychological state of the users on social networks. The posted images and tweets data from users with/without depression on Twitter have been collected and used for depression detection. CNN-based classifier and Bert are applied to extract the deep features from the pictures and text posted by users respectively. Then visual and textual features are combined to reflect the emotional expression of users. Finally our system classifies the users with depression and normal users through a neural network and the analysis report is generated automatically.

MOTIVATION: MicroRNA (miRNA) therapeutics is becoming increasingly important. However, aberrant expression of miRNAs is known to cause drug resistance and can become an obstacle for miRNA-based therapeutics. At present, little is known about associations between miRNA and drug resistance and there is no computational tool available for predicting such association relationship. Since it is known that miRNAs can regulate genes that encode specific proteins that are keys for drug efficacy, we propose here a computational approach, called GCMDR, for finding a three-layer latent factor model that can be used to predict miRNA-drug resistance associations. RESULTS: In this paper, we discuss how the problem of predicting such associations can be formulated as a link prediction problem involving a bipartite attributed graph. GCMDR makes use of the technique of graph convolution to build a latent factor model, which can effectively utilize information of high-dimensional attributes of miRNA/drug in an end-to-end learning scheme. In addition, GCMDR also learns graph embedding features for miRNAs and drugs. We leveraged the data from multiple databases storing miRNA expression profile, drug substructure fingerprints, gene ontology and disease ontology. The test for performance shows that the GCMDR prediction model can achieve AUCs of 0.9301 ± 0.0005, 0.9359 ± 0.0006 and 0.9369 ± 0.0003 based on 2-fold, 5-fold and 10-fold cross validation, respectively. Using this model, we show that the associations between miRNA and drug resistance can be reliably predicted by properly introducing useful side information like miRNA expression profile and drug structure fingerprints. AVAILABILITY AND IMPLEMENTATION: Python codes and dataset are available at https://github.com/yahuang1991polyu/GCMDR/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Long noncoding RNAs (lncRNAs) is an important class of non-protein coding RNAs. They have recently been found to potentially be able to act as a regulatory molecule in some important biological processes. MicroRNAs (miRNAs) have been confirmed to be closely related to the regulation of various human diseases. Recent studies have suggested that lncRNAs could interact with miRNAs to modulate their regulatory roles. Hence, predicting lncRNA-miRNA interactions are biologically significant due to their potential roles in determining the effectiveness of diagnostic biomarkers and therapeutic targets for various human diseases. For the details of the mechanisms to be better understood, it would be useful if some computational approaches are developed to allow for such investigations. As diverse heterogeneous datasets for describing lncRNA and miRNA have been made available, it becomes more feasible for us to develop a model to describe potential interactions between lncRNAs and miRNAs. In this work, we present a novel computational approach called LMNLMI for such purpose. LMNLMI works in several phases. First, it learns patterns from expression, sequences and functional data. Based on the patterns, it then constructs several networks including an expression-similarity network, a functional-similarity network, and a sequence-similarity network. Based on a measure of similarities between these networks, LMNLMI computes an interaction score for each pair of lncRNA and miRNA in the database. The novelty of LMNLMI lies in the use of a network fusion technique to combine the patterns inherent in multiple similarity networks and a matrix completion technique in predicting interaction relationships. Using a set of real data, we show that LMNLMI can be a very effective approach for the accurate prediction of lncRNA-miRNA interactions.

Human immunodeficiency virus 1 (HIV-1) protease (PR) plays a crucial role in the maturation of the virus. The study of substrate specificity of HIV-1 PR as a new endeavor strives to increase our ability to understand how HIV-1 PR recognizes its various cleavage sites. To predict HIV-1 PR cleavage sites, most of the existing approaches have been developed solely based on the homogeneity of substrate sequence information with supervised classification techniques. Although efficient, these approaches are found to be restricted to the ability of explaining their results and probably provide few insights into the mechanisms by which HIV-1 PR cleaves the substrates in a site-specific manner. In this work, a coevolutionary pattern-based prediction model for HIV-1 PR cleavage sites, namely EvoCleave, is proposed by integrating the coevolving information obtained from substrate sequences with a linear SVM classifier. The experiment results showed that EvoCleave yielded a very promising performance in terms of ROC analysis and f-measure. We also prospectively assessed the biological significance of coevolutionary patterns by applying them to study three fundamental issues of HIV-1 PR cleavage site. The analysis results demonstrated that the coevolutionary patterns offered valuable insights into the understanding of substrate specificity of HIV-1 PR.

Recent research has achieved impressive results in single-turn dialogue modelling. In the multi-turn setting, however, current models are still far from satisfactory. One major challenge is the frequently occurred coreference and information omission in our daily conversation, making it hard for machines to understand the real intention. In this paper, we propose rewriting the human utterance as a pre-process to help multi-turn dialgoue modelling. Each utterance is first rewritten to recover all coreferred and omitted information. The next processing steps are then performed based on the rewritten utterance. To properly train the utterance rewriter, we collect a new dataset with human annotations and introduce a Transformer-based utterance rewriting architecture using the pointer network. We show the proposed architecture achieves remarkably good performance on the utterance rewriting task. The trained utterance rewriter can be easily integrated into online chatbots and brings general improvement over different domains.

This paper presents a Generative Adversarial Network (GAN) to model multiturn dialogue generation, which trains a latent hierarchical recurrent encoder-decoder simultaneously with a discriminative classifier that make the prior approximate to the posterior. Experiments show that our model achieves better results.

Possible drug side-effects (SEs) are usually verified by many years of repeated clinical trials. Despite the effort, some drugs are still expected to cause adverse reactions in some patients. To better predict drug SEs without having to go through the laborious processes of testing and re-testing, machine learning (ML) techniques are more and more used to uncovered patterns in drug data for such purpose. Most existing such techniques are black-box techniques. Since correlations between sub-structures involving multiple variables may exist, these techniques may not always work well. For ML techniques to be effective, they should be accurate, efficient and the patterns they discover should be interpretable. Towards these goals, we have developed a second-order association discovering (SOAD) algorithm for SE prediction. Given a set of drug data for training, the SOAD algorithm can discover SO associations between multiple drug sub-structures and multiple SEs in drug data for the purpose of predicting the SEs. SOAD performs its tasks by first making use of a residual measure to test the significance of occurrence of a chemical sub-structure within a drug and the SE of the drug. Once an association is established between a sub-structure and a SE, we test if two or more such sub-structures are significantly associated with a SE. Based on such second-order associations, we derive from them a set of “informative” SO patterns so that the SEs of new unseen drugs can be predicted based on the frequency of appearance of such patterns. To ensure interpretability of the SE discovery process, we make use of the Bayesian to predict if certain SO relationship in a drug may be related to a certain side-effect. Based on the experimental results, SOAD is found to be very promising.

The rapid development of techniques results in a growing diversity of social network data which require for analysis. Therefore, the deeper understanding of latent knowledge representing the social network data needs learning by combining the insights obtained from multiple, diverse networks carrying heterogeneous information featuring the interrelationship between vertices. In this manuscript, we propose a novel deepmodel- based approach to learn latent structural representation from multi-domain social network data. The algorithm, which we call Deep Multiple Networks Fusion (DMNF), is able to discover an aggregated deep representation, by taking into consideration multiple networks, which represent heterogeneous information carried by the social network data. To perform the task, DMNF first constructs a network representing the total degree of interrelationship between pairwise vertices by utilizing a fusion method to compute such degree taking into consideration heterogeneous information embedded in the network data, e.g., node connection, and attribute relativity. Given the fused network data, DMNF attempts to learn the latent network representation making use of a deep neural network model. Such learned representation is able to reveal the latent structure, e.g., social communities, and clusters in the social network. DMNF has been tested with two sets of real social network data and compared with several prevalent approaches to network community detection. The experimental results show that the latent representation found by DMNF may match well with the ground-truth communities and DMNF is able to outperform the state-of-the-art approaches to detecting social network communities.