The vast chemical space, encompassing virtually innumerable molecules and materials, presents both immense opportunities and significant challenges. The design and discovery of novel drugs and functional materials may be viewed as a search within this space; however, the sheer scale of potential candidates renders exhaustive exploration infeasible. To address this, we introduce Chemical Mapper, a framework that integrates topological data analysis with deep learning to enable the visual exploration and analysis of chemical latent spaces. At its core, Chemical Mapper employs mapper, a widely used tool in topological data analysis, to investigate the organizational principles of chemical latent spaces defined by molecular representations learned by geometric deep learning models. In doing so, Chemical Mapper not only highlights groups of molecular representations but also uncovers the relationships among them through linkages and branching structures. Our results show that Chemical Mapper reveals intrinsic patterns associated with molecular scaffolds, functional groups, and chemical properties, as well as the structural and functional evolutions of the molecules.

MOTIVATION: Accurately predicting anticancer drug response is a central challenge in precision oncology. Existing computational methods, although valuable, often depend on pairwise molecular descriptors or limited graph-based encodings that cannot fully capture the complexity of molecular structures or their interactions with cellular states. These constraints hinder their robustness and generalization across diverse drugs and biological contexts, underscoring the need for more expressive frameworks. RESULTS: To address this gap, we propose MolDr, a topological deep learning framework that represents molecules as multiscale simplicial complexes and propagates information across higher-order structures. By integrating these molecular representations with cellular profiles, MolDr unifies chemical topology and biological context within a single predictive model. Comprehensive experiments show that MolDr consistently outperforms or matches state-of-the-art baselines across multiple benchmarks. It achieves stronger accuracy and robustness on continuous drug response tasks, while also generalizing effectively to discrete classification settings. Moreover, sensitivity analysis confirms the benefit of incorporating multiple topological scales, further supporting the importance of higher-order representations. Together, these results demonstrate that MolDr delivers reliable performance across heterogeneous pharmacogenomic scenarios and highlight the promise of topological modeling for advancing drug response prediction. AVAILABILITY: Source code freely available at https://github.com/CS-BIO/MolDr.

Spatial transcriptomics technologies enable the generation of gene expression profiles while retaining spatial coordinates. Identifying spatial domains and spatially variable genes (SVGs) are crucial tasks in spatial transcriptomics, offering valuable insights into biological functions. However, a deep learning framework that integrates SVGs detection with spatial domain identification is still lacking. In this study, we propose a multi-task ensemble analysis framework for spatial transcriptomics, named SpaMCI-DL, which adopts multi-constrained interpretable deep learning to jointly perform SVGs detection and spatial domain identification. SpaMCI-DL first employs a graph convolutional autoencoder to identify spatial domains by incorporating binary and graph structural constraints. Subsequently, based on the learned spatial domains, SpaMCI-DL utilizes a gradients-based method with multi-scale constraints to detect SVGs, enhancing the interpretability and biological relevance of the results. Comparative evaluations against state-of-the-art methods on five spatial transcriptomics datasets, spanning diverse species and tissues, demonstrate that SpaMCI-DL achieves superior performance in both spatial domain identification and SVGs detection. The code are available at https://github.com/liangxiao-cs/SpaMCI-DL.

Geometric deep learning (GDL) models have demonstrated a great potential for the analysis of non-Euclidian data. They are developed to incorporate the geometric and topological information of non-Euclidian data into the end-to-end deep learning architectures. Motivated by the recent success of discrete Ricci curvature in graph neural network (GNNs), we propose TorGNN, an analytic Torsion enhanced Graph Neural Network model. The essential idea is to characterize graph local structures with an analytic torsion based weight formula. Mathematically, analytic torsion is a topological invariant that can distinguish spaces which are homotopy equivalent but not homeomorphic. In our TorGNN, for each edge, a corresponding local simplicial complex is identified, then the analytic torsion (for this local simplicial complex) is calculated, and further used as a weight (for this edge) in message-passing process. Our TorGNN model is validated on link prediction tasks from sixteen different types of networks and node classification tasks from four types of networks. It has been found that our TorGNN can achieve superior performance on both tasks, and outperform various state-of-the-art models. This demonstrates that analytic torsion is a highly efficient topological invariant in the characterization of graph structures and can significantly boost the performance of GNNs.

Accurate and efficient prediction of polymers properties is crucial for polymer design. Recently, data-driven artificial intelligence (AI) models have demonstrated great promise in polymers property analysis. Even with the great progresses, a pivotal challenge in all the AI-driven models remains to be the effective representation of molecules. Here we introduce Multi-Cover Persistence (MCP)-based molecular representation and featurization for the first time. Our MCP-based polymer descriptors are combined with machine learning models, in particular, Gradient Boosting Tree (GBT) models, for polymers property prediction. Different from all previous molecular representation, polymer molecular structure and interactions are represented as MCP, which utilizes Delaunay slices at different dimensions and Rhomboid tiling to characterize the complicated geometric and topological information within the data. Statistic features from the generated persistent barcodes are used as polymer descriptors, and further combined with GBT model. Our model has been extensively validated on polymer benchmark datasets. It has been found that our models can outperform traditional fingerprint-based models and has similar accuracy with geometric deep learning models. In particular, our model tends to be more effective on large-sized monomer structures, demonstrating the great potential of MCP in characterizing more complicated polymer data. This work underscores the potential of MCP in polymer informatics, presenting a novel perspective on molecular representation and its application in polymer science.

Drug-drug interaction (DDI) can trigger many adverse effects in patients and has emerged as a threat to medicine and public health. Therefore, it is important to predict potential drug interactions since it can provide combination strategies of drugs for systematic and effective treatment. Existing deep learning-based methods often rely on DDI functional networks, or use them as an important part of the model information source. However, it is difficult to discover the interactions of a new drug. To address the above limitations, we propose a geometric molecular graph representation learning model (Mol-DDI) for DDI prediction based on the basic assumption that structure determines function. Mol-DDI only considers the covalent and non-covalent bond information of molecules, then it uses the pre-training idea of large-scale models to learn drug molecular representations and predict drug interactions during the fine-tuning process. Experimental results show that the Mol-DDI model outperforms others on the three datasets and performs better in predicting new drug interaction experiments.

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.

single-cell RNA-sequencing (scRNA-seq) technology can reveal cellular heterogeneity with high throughput and resolution, facilitating the profiling of single-cell transcriptomes. However, due to some experimental factors, a large number of missing values are generated in scRNA-seq data, which are called dropout events, and this phenomenon affects the downstream analysis. Imputation is an effective denoising method, but existing imputation methods still face a huge challenge: lack of interpretability. In this study, we propose single-cell Self-Attention Generative Adversarial Networks(scSAGAN), a semi-supervised imputation method for scRNA-seq data. scSAGAN mainly uses Semi-Supervised Learning (SSL) and Probabilistic Latent Semantic Analysis (PLSA), which can not only learn the potential characteristics of different types of cells but explain their imputation behavior. In clustering experiments, scSAGAN exhibits better clustering performance than all baselines on 7 datasets. Next, we interpret the imputation behavior of scSAGAN on datasets such as Alzheimer’s disease and find causative genes associated with the corresponding datasets. scSAGAN is currently an open-source method, available at https://github.com/zehaoxiongl23/scSAGAN.

Accumulated studies have found that miRNAs are in charge of many complex diseases such as cancers by modulating gene expression. Predicting miRNA-target interactions is beneficial for uncovering the crucial roles of miRNAs in regulating target genes and the progression of diseases. The emergence of large-scale genomic and biological data as well as the recent development in heterogeneous networks provides new opportunities for miRNA target identification. Compared with conventional methods, computational methods become a decent solution for high efficiency. Thus, designing a method that could excavate valid information from the heterogeneous network and gene sequences is in great demand for improving the prediction accuracy. In this study, we proposed a graph-based model named MRMTI for the prediction of miRNA-target interactions. MRMTI utilized the multi-relation graph convolution module and the Bi-LSTM module to incorporate both network topology and sequential information. The learned embeddings of miRNAs and genes were then used to calculate the prediction scores of miRNA-target pairs. Comparisons with other state-of-the-art graph embedding methods and existing bioinformatic tools illustrated the superiority of MRMTI under multiple criteria metrics. Three variants of MRMTI implied the positive effect of multi-relation. The experimental results of case studies further demonstrated the prominent ability of MRMTI in predicting novel associations.

MicroRNAs (miRNAs) play a significant role in regulating gene transcription and tend to act in a combinatorial way, which provides great insights to explore disease-related miRNA pairs or modules for comprehending the synergistic roles of miRNAs in complex diseases. As wet experiments are often laborious and costly, computational methods offer great convenience for predicting potential associations between miRNAs and diseases. Existing methods focus on either the ‘one miRNA-one disease’ paradigm, or merely the synergetic miRNA network about specific diseases, which may lead to the incomplete understanding of the synergistic effect of miRNAs on the pathogenesis of complex diseases. In this work, we present a novel tensor-based framework, named GraphTF1, to predict disease-associated miRNA-miRNA pairs. GraphTF exploits graph attention network to effectively capture node features over multi-source biological network. Then, the learned miRNA and disease representations are used to reconstruct the association tensor for predicting potential disease-associated miRNA-miRNA pairs. Empirical results showed that the proposed method outperformed all other state-of-the-art methods under five-fold cross-validation. Robustness experiments also indicated the stability of GraphTF. Moreover, case studies for Breast Neoplasms and Lung Neoplasms further demonstrated the effectiveness of GraphTF in identifying potential disease-related miRNA-miRNA pairs.

MOTIVATION: In recent years, a growing number of studies have proved that microRNAs (miRNAs) play significant roles in the development of human complex diseases. Discovering the associations between miRNAs and diseases has become an important part of the discovery and treatment of disease. Since uncovering associations via traditional experimental methods is complicated and time-consuming, many computational methods have been proposed to identify the potential associations. However, there are still challenges in accurately determining potential associations between miRNA and disease by using multisource data. RESULTS: In this study, we develop a Multi-view Multichannel Attention Graph Convolutional Network (MMGCN) to predict potential miRNA-disease associations. Different from simple multisource information integration, MMGCN employs GCN encoder to obtain the features of miRNA and disease in different similarity views, respectively. Moreover, our MMGCN can enhance the learned latent representations for association prediction by utilizing multichannel attention, which adaptively learns the importance of different features. Empirical results on two datasets demonstrate that MMGCN model can achieve superior performance compared with nine state-of-the-art methods on most of the metrics. Furthermore, we prove the effectiveness of multichannel attention mechanism and the validity of multisource data in miRNA and disease association prediction. Case studies also indicate the ability of the method for discovering new associations.

MOTIVATION: Protein kinases have been the focus of drug discovery research for many years because they play a causal role in many human diseases. Understanding the binding profile of kinase inhibitors is a prerequisite for drug discovery, and traditional methods of predicting kinase inhibitors are time-consuming and inefficient. Calculation-based predictive methods provide a relatively low-cost and high-efficiency approach to the rapid development and effective understanding of the binding profile of kinase inhibitors. Particularly, the continuous improvement of network pharmacology methods provides unprecedented opportunities for drug discovery, network-based computational methods could be employed to aggregate the effective information from heterogeneous sources, which have become a new way for predicting the binding profile of kinase inhibitors. RESULTS: In this study, we proposed a network-based influence deep diffusion model, named IDDkin, for enhancing the prediction of kinase inhibitors. IDDkin uses deep graph convolutional networks, graph attention networks and adaptive weighting methods to diffuse the effective information of heterogeneous networks. The updated kinase and compound representations are used to predict potential compound-kinase pairs. The experimental results show that the performance of IDDkin is superior to the comparison methods, including the state-of-the-art kinase inhibitor prediction method and the classic model widely used in relationship prediction. In experiments conducted to verify its generalizability and in case studies, the IDDkin model also shows excellent performance. All of these results demonstrate the powerful predictive ability of the IDDkin model in the field of kinase inhibitors. AVAILABILITY AND IMPLEMENTATION: Source code and data can be downloaded from https://github.com/CS-BIO/IDDkin. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Small molecule(SM) drugs can affect the expression of miRNAs, which plays crucial roles in many important biological processes. The chemical structure and clinical information of small molecule can simultaneously incorporate information such as anatomical distribution, therapeutic effects and structural characteristics. It is necessary to develop a novel model that incorporates small molecule chemical structure and clinical information to reveal the unknown small molecule-miRNA associations. In this study, we developed a new framework based on non-negative matrix factorization, called SMANMF, to discover the potential small molecules-miRNAs associations. First, the functional similarity of two miRNAs can be obtained by computing the overlap of the target gene sets in which the miRNAs interact together, and we integrated two types of small molecule similarities, including chemical similarity and clinical similarity. Then, we utilized a non-negative matrix factorization model to discover the unknown relationship between small molecules and miRNAs. The evaluation results indicate that our model can achieve superior prediction performance compared with previous approaches in 5-fold cross-validation. At the same time, the results of case studies also reveal that the SMANMF model has good predictive performance for predicting the potential association between small molecules and miRNAs.