Recent advances in spatial molecular imaging technologies have enabled gene expression profiling alongside high-resolution imaging, providing unprecedented opportunities to resolve molecular heterogeneity at subcellular resolution. However, these technologies fail to fully capture cellular characteristics due to the limited number of genes they can detect, which hinder downstream analysis. Spatial imaging data provide high-resolution and fine-grained morphology information, developing computational methods that effectively integrate image features with transcriptomic profiles is crucial for enabling comprehensive subcellular data analysis. In this study, we present SIMMT, an image-enhanced multi-modal contrastive transformer framework for identifying spatial domains and enhancing subcellular data. In the framework, we design a dual transformer architecture to learn multi-modal representations for cells by modeling transcriptomics and morphological images respectively. To fully capture modality interactions within spatial contexts, we introduce a contrastive learning module that enhances cell representation by aligning tissue morphology and gene expression at the cell level. We tested SIMMT on subcellular spatial transcriptomics datasets from human lung cancer tissue, mouse brain tissue, human colorectal cancer tissue, and human ovarian cancer tissue. The results demonstrated that SIMMT consistently outperformed state-of-the-art methods in spatial clustering and gene expression pattern analysis. Our method also effectively demonstrated its ability to identify tumor spatial heterogeneity and uncover potential gene biomarkers in the human bronchiolar adenoma (BA) dataset.

In this work, we aim to reconstruct the 3D shape of an indoor scene from a single view, which includes multiple objects and the background. This task is challenging for existing methods since those instances of indoor scenes regularly occlude each other and contain diverse topologies. To address this, we propose a novel framework, ISDNet, to adaptively separate mixed instances and perform topology-aware reconstruction. Specifically, Specifically, ISDNet consists of two cascaded subnetworks: an instance separation module (ISM) and an instance deformation module (IDM). The ISM learns to separate occluded objects through stepwise sampling, inferring clean features for each instance. On the basis of these features, IDM generates an instance-topology-aware template and deforms it with learned offsets to reconstruct detailed geometry. Quantitative and qualitative experiments on the SUNRGB-D and 3D-FRONT datasets demonstrate that ISDNet outperforms the state-of-theart methods in terms of local details and overall shapes.

Existing point cloud completion methods rely on extracting latent codes from a partial point cloud to reconstruct a complete structure. However, the complexity of the partial point clouds, making the completion results of such methods less satisfactory, especially in long-distance (away from partial point cloud) areas. To tackle this challenge, we propose a point cloud completion network via heuristic structure growing (HSG-Net), which progressively completes the close-distance structure through an iterative heuristic structure growth strategy. Particularly, a novel data preprocessing (DP) method is proposed to obtain ground truth (GT) with specific structural integrity, guiding the network to learn close-distance structural information. In addition, the proposed consistency constraint displacement module (CCDM) is employed to fulfill structure growth, and a feature memory module (FMM) further enhances the quality of the grown structure. Furthermore, a proposed local information generator is used to further refine the structure-grown point cloud, fetching the final result. Extensive quantitative and qualitative results demonstrate that our HSG-Net outperforms the state-of-the-art methods.

Enhancers play a critical role in gene expression regulation. However, their accurate prediction remains a significant challenge due to the limited feature information provided by sequence semantics. To address this issue, we propose a novel multimodal framework, termed iEnhancer-Fusion, for enhancer identification and classification. The proposed model integrates two complementary modalities: DNA sequence features extracted using DNABERT-2, and DNA breathing features captured through a hybrid network comprising convolutional layer and Multi-Head Attention mechanism. These heterogeneous features are further fused via a Cross-Attention mechanism, enabling deep interaction between modalities and effectively overcoming the representational limitations of sequence-only models. Comparative experiments against seven representative enhancer prediction methods across two tasks demonstrate that iEnhancer-Fusion achieves superior performance across all key evaluation metrics. Specifically, in Task 1, the model achieves average ACC, MCC, and AUC scores of$82.70 \%, 65.52 \%$, and 87.35 %, respectively; in Task 2, the average scores for ACC, MCC, and AUC are$93.10 \%, 86.88 \%$, and 97.54 %, respectively.

Recent advances in spatial transcriptomics have enabled subcellular-resolution profiling of gene expression, offering unprecedented opportunities to investigate intracellular architecture and local microenvironmental interactions. Graph neural networks (GNNs) have shown great promise in modeling spatial transcriptomics data. However, existing GNN-based methods primarily focus on low-frequency signals, overlooking high-frequency signals critical for resolving transcriptional differences across subcellular compartments and cell boundaries. This limits their ability to characterize fine-grained structural and functional heterogeneity within tissues, hindering accurate spatial domain identification. In this study, we propose HiFi-ST, a High-Frequency-Aware Graph Integration framework for subcellular spatial transcriptomics. HiFi-ST employs a high-pass filter to extract high-frequency transcriptional differences, which are then integrated with spatial contexts through a transformer-based architecture. A contrastive learning module is designed to enhance cell representation by aligning spatial organization with transcriptional heterogeneity. Comprehensive experiments on subcellular datasets demonstrated that HiFi-ST consistently outperformed six state-of-the-art methods in spatial clustering, gene expression enhancement, and niche identification.

3D Gaussian Splatting (3D-GS) has exhibited impressive progress in novel view synthesis. When given the sparse views, its performance degrades severely, causing many problems like novel views collapse and excessive floaters. Many recent methods take into account fitting input views, inferring missing scene information and optimizing the final scene representation, all through a single stage. After revisiting the task, we propose a novel framework, Task-Disentangled 3D Gaussian Splatting, abbreviated to TD-GS. It splits the sparse views synthesis task into two subtasks: (i) Dense Generation. (ii) Enhanced Synthesis. In the subtask of Dense Generation, we estimate dense views from sparse input. Then in the subtask of Enhanced Synthesis, both the dense views and the sparse input participate in the training of Gaussians to obtain the final Gaussian representation of the scene. In the process of completing the first subtask, we carefully design Gaussian Cloud Denoising to directly edit 3D Gaussians. Also, we introduce two regularization methods to guide the geometric optimization towards an optimal solution. The purpose is to estimate more reliable outputs. Many experiments have validated that our TD-GS outperforms other state-of-the-art methods.

Scene sketching is to convert a scene into a simplified, abstract representation that captures the essential elements and composition of the original scene. It requires a semantic understanding of the scene and consideration of different regions within the scene. Since scenes often contain diverse visual information across various regions, such as foreground objects, background elements, and spatial divisions, dealing with these different regions poses unique difficulties. In this paper, we define a sketch as some sets of Bézier curves because of their smooth and versatile characteristics. We optimize different regions of input scene in multiple rounds. In each optimization round, the strokes sampled from the next region can seamlessly be integrated into the sketch generated in the previous optimization round. We propose an additional stroke initialization method to ensure the integrity of the scene and the convergence of optimization. A novel CLIP-based Semantic Loss and a VGG-based Feature Loss are utilized to guide our multi-round optimization. Extensive experimental results on the quality and quantity of the generated sketches confirm the effectiveness of our method.

Semantic Scene Completion (SSC) aims to generate a complete semantic scene from an incomplete input. Existing approaches often employ dense network architectures with a high parameter count, leading to increased model complexity and resource demands. To address these limitations, we propose RWKV-PCSSC, a lightweight point cloud semantic scene completion network inspired by the Receptance Weighted Key Value (RWKV) mechanism. Specifically, we introduce a RWKV Seed Generator (RWKV-SG) module that can aggregate features from a partial point cloud to produce a coarse point cloud with coarse features. Subsequently, the point-wise feature of the point cloud is progressively restored through multiple stages of the RWKV Point Deconvolution (RWKV-PD) modules. By leveraging a compact and efficient design, our method achieves a lightweight model representation. Experimental results demonstrate that RWKV-PCSSC reduces the parameter count by 4.18× and improves memory efficiency by 1.37× compared to state-of-the-art methods PointSSC[51]. Furthermore, our network achieves state-of-the-art performance on established indoor (SSC-PC, NYUCAD-PC) and outdoor (PointSSC) scene dataset, as well as on our proposed datasets (NYUCAD-PC-V2, 3D-FRONT-PC).

Cell type annotation is crucial in single-cell RNA sequencing data analysis because it enables significant biological discoveries and deepens our understanding of tissue biology. Given the high-dimensional and highly sparse nature of single-cell RNA sequencing data, most existing annotation tools focus on highly variable genes to reduce dimensionality and computational load. However, this approach inevitably results in information loss, potentially weakening the model's generalization performance and adaptability to novel datasets. To mitigate this issue, we developed scTrans, a single cell Transformer-based model, which employs sparse attention to utilize all non-zero genes, thereby effectively reducing the input data dimensionality while minimizing information loss. We validated the speed and accuracy of scTrans by performing cell type annotation on 31 different tissues within the Mouse Cell Atlas. Remarkably, even with datasets nearing a million cells, scTrans efficiently perform cell type annotation in limited computational resources. Furthermore, scTrans demonstrates strong generalization capabilities, accurately annotating cells in novel datasets and generating high-quality latent representations, which are essential for precise clustering and trajectory analysis.

The advent of single-cell RNA sequencing (scRNA-seq) has offering unprecedented insights at the single-cell level. This groundbreaking technology has opened new pathways for understanding cellular diversity and revealing novel insights into disease mechanisms. However, the analysis of scRNA-seq data is challenging, primarily due to dropout events caused by technical noise. Developing effective imputation methods is crucial for the reliable and informative analysis of scRNA-seq data. While deep learning-based approaches have been proposed for scRNA-seq data imputation, they often fall short of optimal performance, especially in identifying rare cell types. Here we propose a novel self-supervised deep learning model named scGANCL for scRNA-seq data imputation. scGANCL combines bidirectional generative adversarial network (BiGAN) with contrastive learning (CL) to enhance imputation performance. To fully exploit gene expression profiles, a contrastive learning module is introduced to enhance the representation learning of cells by minimizing the discrepancy between the distributions of real and generated data. Comprehensive experiments have been conducted on ten simulated and seven real datasets to validate scGANCL's effectiveness. The results demonstrated scGANCL consistently outperformed seven state-of-the-art methods across various downstream tasks. Ablation studies further validated the contribution of each component to the overall performance of the model.

MicroRNAs (miRNAs) mediates gene expression regulation by targeting specific messenger RNAs (mRNAs) in the cytoplasm. They can function as both tumor suppressors and oncogenes depending on the specific miRNA and its target genes. Detecting miRNA-target interactions (MTIs) is critical for unraveling the complex mechanisms of gene regulation and promising towards RNA therapy for cancer. There is currently a lack of MTIs prediction methods that simultaneously perform feature learning from heterogeneous gene regulatory network (GRN) and genomic sequences. To improve the prediction performance of MTIs, we present a novel transformer-based multi-view feature learning method – MiRGraph, which consists of two main modules for learning the sequence-based and GRN-based feature embedding. For the former, we utilize the mature miRNA sequences and the complete 3'UTR sequence of the target mRNAs to encode sequence features using a hybrid transformer and convolutional neural network (CNN) (TransCNN) architecture. For the latter, we utilize a heterogeneous graph transformer (HGT) module to extract the relational and structural information from the GRN consisting of miRNA-miRNA, gene-gene and miRNA-target interactions. The TransCNN and HGT modules can be learned end-to-end to predict experimentally validated MTIs from MiRTarBase. MiRGraph outperforms existing methods in not only recapitulating the true MTIs but also in predicting strength of the MTIs based on the in-vitro measurements of miRNA transfections. In a case study on breast cancer, we identified plausible target genes of an oncomir.

Speech-driven facial video is a one-to-many mapping problem where each input audio can have multiple plausible facial outputs, leading to overly smooth facial movements results. To overcome this problem, we introduce discrete motion priors to reduce the uncertainty in facial movements and enhance realism. Simultaneously through adversarial training, we achieve domain adaptation, mapping facial movements to a low-dimensional grid space to enhance adaptability to external audio. Following that we establish an explicit connection between facial movements region and spatial regions using a spatial position attention to improve the precision of dynamic portrait modeling. Finally we expedite the generation of speaking face videos using a grid-based dynamic neural radiation field to render the head and torso separately. Experimental results demonstrate that our approach outperforms previous methods, yielding more extensive and realistic speech-driven facial video.

Single-cell RNA sequencing (scRNA-seq) technology is utilized to analyze cellular heterogeneity, perform cellular-level biological research and derive novel insights from complex cellular systems. However, the raw scRNA-seq data is not directly suitable for downstream task analysis due to its high variability, sparsity and dimensionality. Therefore, in this study, we propose a new self-supervised framework based on siamese representation learning, named scSRL which can fully explore the intrinsic properties of cells by maximizing the similarity between positive pairs. These positive pairs are constructed by multiple data augmentation operations to further increase data diversity and better learn latent representation. Moreover, our method employs a gradient stopping strategy to mitigate collapsing in the siamese network. It is worth noting that the scSRL focuses on aggregating cells with similar functions without introducing negative samples, which can avoid additional computational cost. Finally, We evaluated scSRL on 10 real datasets for downstream tasks such as clustering, classification and visualization, and it consistently exhibited outstanding performance in all these fundamental tasks. Meanwhile, we did pseudotime inference experiments in two embryonic development datasets, and the scSRL model can accurately reconstruct cell trajectory and describe cell developmental process. scSRL is currently an open-source method, available at https://github.com/zysun17/scSRL.

Point clouds obtained from 3D scanning are typically incomplete, noisy, and sparse. Previous completion methods aim to generate complete point clouds, while taking into account the densification of point clouds, filling small holes, and proximity-to-surface, all through a single network. After revisiting the task, we propose SDNet, which disentangles the task based on the spatial characteristics of point clouds and formulates two sub-networks, a Dense Refiner and a Missing Generator. Given a partial input, the Dense Refiner produces a dense and clean point cloud, as a more reliable partial surface, which assists the Missing Generator to better infer the remaining point cloud structure. To promote the alignment and interaction across these two modules, we propose a Cross Fusion Unit with designed Non-Symmetrical Cross Transformers to capture geometric relationships between partial and missing regions, contributing to a complete, dense and well-aligned output. Extensive quantitative and qualitative results demonstrate that our method outperforms the state-of-the-art methods.

MOTIVATION: Recent advances in spatial transcriptomics technologies have enabled gene expression profiles while preserving spatial context. Accurately identifying spatial domains is crucial for downstream analysis and it requires the effective integration of gene expression profiles and spatial information. While increasingly computational methods have been developed for spatial domain detection, most of them cannot adaptively learn the complex relationship between gene expression and spatial information, leading to sub-optimal performance. RESULTS: To overcome these challenges, we propose a novel deep learning method named Spatial-MGCN for identifying spatial domains, which is a Multi-view Graph Convolutional Network (GCN) with attention mechanism. We first construct two neighbor graphs using gene expression profiles and spatial information, respectively. Then, a multi-view GCN encoder is designed to extract unique embeddings from both the feature and spatial graphs, as well as their shared embeddings by combining both graphs. Finally, a zero-inflated negative binomial decoder is used to reconstruct the original expression matrix by capturing the global probability distribution of gene expression profiles. Moreover, Spatial-MGCN incorporates a spatial regularization constraint into the features learning to preserve spatial neighbor information in an end-to-end manner. The experimental results show that Spatial-MGCN outperforms state-of-the-art methods consistently in several tasks, including spatial clustering and trajectory inference.

MOTIVATION: Single-cell RNA-sequencing (scRNA-seq) is widely used to reveal cellular heterogeneity, complex disease mechanisms and cell differentiation processes. Due to high sparsity and complex gene expression patterns, scRNA-seq data present a large number of dropout events, affecting downstream tasks such as cell clustering and pseudo-time analysis. Restoring the expression levels of genes is essential for reducing technical noise and facilitating downstream analysis. However, existing scRNA-seq data imputation methods ignore the topological structure information of scRNA-seq data and cannot comprehensively utilize the relationships between cells. RESULTS: Here, we propose a single-cell Graph Contrastive Learning method for scRNA-seq data imputation, named scGCL, which integrates graph contrastive learning and Zero-inflated Negative Binomial (ZINB) distribution to estimate dropout values. scGCL summarizes global and local semantic information through contrastive learning and selects positive samples to enhance the representation of target nodes. To capture the global probability distribution, scGCL introduces an autoencoder based on the ZINB distribution, which reconstructs the scRNA-seq data based on the prior distribution. Through extensive experiments, we verify that scGCL outperforms existing state-of-the-art imputation methods in clustering performance and gene imputation on 14 scRNA-seq datasets. Further, we find that scGCL can enhance the expression patterns of specific genes in Alzheimer's disease datasets. AVAILABILITY AND IMPLEMENTATION: The code and data of scGCL are available on Github: https://github.com/zehaoxiong123/scGCL. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Precise identification of target sites of RNA-binding proteins (RBP) is important to understand their biochemical and cellular functions. A large amount of experimental data is generated by in vivo and in vitro approaches. The binding preferences determined from these platforms share similar patterns but there are discernable differences between these datasets. Computational methods trained on one dataset do not always work well on another dataset. To address this problem which resembles the classic "domain shift" in deep learning, we adopted the adversarial domain adaptation (ADDA) technique and developed a framework (RBP-ADDA) that can extract RBP binding preferences from an integration of in vivo and vitro datasets. Compared with conventional methods, ADDA has the advantage of working with two input datasets, as it trains the initial neural network for each dataset individually, projects the two datasets onto a feature space, and uses an adversarial framework to derive an optimal network that achieves an optimal discriminative predictive power. In the first step, for each RBP, we include only the in vitro data to pre-train a source network and a task predictor. Next, for the same RBP, we initiate the target network by using the source network and use adversarial domain adaptation to update the target network using both in vitro and in vivo data. These two steps help leverage the in vitro data to improve the prediction on in vivo data, which is typically challenging with a lower signal-to-noise ratio. Finally, to further take the advantage of the fused source and target data, we fine-tune the task predictor using both data. We showed that RBP-ADDA achieved better performance in modeling in vivo RBP binding data than other existing methods as judged by Pearson correlations. It also improved predictive performance on in vitro datasets. We further applied augmentation operations on RBPs with less in vivo data to expand the input data and showed that it can improve prediction performances. Lastly, we explored the predictive interpretability of RBP-ADDA, where we quantified the contribution of the input features by Integrated Gradients and identified nucleotide positions that are important for RBP recognition.

MOTIVATION: MircroRNAs (miRNAs) regulate target genes and are responsible for lethal diseases such as cancers. Accurately recognizing and identifying miRNA and gene pairs could be helpful in deciphering the mechanism by which miRNA affects and regulates the development of cancers. Embedding methods and deep learning methods have shown their excellent performance in traditional classification tasks in many scenarios. But not so many attempts have adapted and merged these two methods into miRNA-gene relationship prediction. Hence, we proposed a novel computational framework. We first generated representational features for miRNAs and genes using both sequence and geometrical information and then leveraged a deep learning method for the associations' prediction. RESULTS: We used long short-term memory (LSTM) to predict potential relationships and proved that our method outperformed other state-of-the-art methods. Results showed that our framework SG-LSTM got an area under curve of 0.94 and was superior to other methods. In the case study, we predicted the top 10 miRNA-gene relationships and recommended the top 10 potential genes for hsa-miR-335-5p for SG-LSTM-core. We also tested our model using a larger dataset, from which 14 668 698 miRNA-gene pairs were predicted. The top 10 unknown pairs were also listed. AVAILABILITY: Our work can be download in https://github.com/Xshelton/SG_LSTM. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Briefings in Bioinformatics online.

Long non-coding RNAs (lncRNAs) have been shown to be involved in multiple biological processes and play critical roles in tumorigenesis. Numerous lncRNAs have been discovered in diverse species, but the functions of most lncRNAs still remain unclear. Meanwhile, their expression patterns and regulation mechanisms are also far from being fully understood. With the advances of high-throughput technologies, the increasing availability of genomic data creates opportunities for deciphering the molecular mechanism and underlying pathogenesis of human diseases. Here, we develop an integrative framework called JONMF to identify lncRNA-mRNA co-expression modules based on the sample-matched lncRNA and mRNA expression profiles. We formulate the module detection task as an optimization problem with joint orthogonal non-negative matrix factorization that could effectively prevent multicollinearity and produce a good modularity interpretation. The constructed lncRNA-mRNA co-expression network and the gene-gene interaction network are used as the network-regularized constraints to improve the module accuracy, while the sparsity constraints are simultaneously utilized to achieve modular sparse solutions. We applied JONMF to human ovarian cancer dataset and the experiment results demonstrate that the proposed method can effectively discover biologically functional co-expression modules, which may provide insights into the function of lncRNAs and molecular mechanism of human diseases.

Predicting the miRNA-target interactions (MTIs) is a critical task for elucidating mechanistic roles of miRNAs in pathophysiology. However, most existing techniques have a higher false positive because the precise miRNA target mechanisms are poorly known. Considering that ensemble methods can take advantage of the complementary knowledge in different methods, we propose an alternative optimization framework, Inferring MiRNA Targets based on Restricted Boltzmann Machines (IMTRBM), to enhance the accuracy of previous prediction results. First, the proposed method directly constructs a weighted MTI network though the results predicted by individual methods and each miRNA target pair is weighted based on the frequency appearing in these results. Second, we transform the miRNA-target prediction problem into a complete bipartite graph model, named restricted Boltzmann machine, and utilize a practical learning procedure to train our model and make predictions. Our results show that the algorithm outperforms individual miRNA-target prediction approach in the number of validated miRNA targets at cutoffs of top list. Moreover, our framework can tolerate the decrease and increase of predicted MTIs and even discover new miRNA targets, which have been a challenge to predict for any individual methods. Finally, for the miRNAs that are not appearing in IMTRBM, we design a new method to supplement IMTRBM based on the intuition that similar miRNAs have similar functions, which also achieves a comparable result. The source code of IMTRBM is available at https://github.com/liuying201705/IMTRBM.