Accurately identifying drug-target interactions (DTIs) is a critical step in drug discovery. While structure-based drug design methods demonstrate impressive docking prediction accuracy, their heavy computational demands and resource-intensive nature make them impractical for directly processing vast chemical spaces containing a large number of compounds. This limitation highlights the need for a computational tool that balance speed and accuracy to rank and filter potential drug candidates efficiently. In contrast, existing ligand-based drug design methods, which learn representations from diverse protein and molecule features, often fail to make consistent predictions on unseen data or external databases, limiting their applicability for ranking and filtering potential drug candidates accurately. To address these challenges, we propose EnsDTI, a novel framework that bridges the gap between structure-based and ligand-based drug design approaches. EnsDTI utilizes a mixture-of-experts architecture to enhance DTI predictions using existing deep learning models and incorporates an inductive conformal predictor to assess prediction quality with confidence scores, ensuring reliability. Experimental results on four widely used benchmark datasets show that EnsDTI consistently achieves high performance in both prediction accuracy and confidence estimation. In addition, its candidate rankings correlate well with actual docking affinities, suggesting its practical utility in drug discovery.

The simultaneous use of multiple medications is a common practice in disease treatment, yet the same drug combination can lead to different effects under varying physiological, pharmacological, or genomic conditions-collectively referred to as the 'context'. Accurately predicting the outcomes of drug combinations across diverse contexts, also known as drug relational learning (DRL), is essential for improving therapeutic efficacy and safety. Despite its importance, existing methods face two major challenges: they are often tailored to specific DRL tasks, lacking generalizability, and they fail to explicitly model the influence of context on drug interactions. This limitation arises because most methods focus primarily on whole-drug compound structures, overlooking the fine-grained atomic-level interactions critical for context-aware predictions. To address these challenges, we propose a novel context-aware hierarchical fusion architecture for DRL. By formulating the problem as the label prediction of drug-drug-context triplets, our approach explicitly models the interaction between drugs by first learning their intrinsic atomic-level interactions and then incorporating context into their embeddings at the atomic level through information fusion. Experiments across diverse tasks-such as synergy prediction, polypharmacy side effect detection, and drug-drug interaction prediction-demonstrate our model's capability to effectively capture context-aware information. Importantly, our method consistently achieves robust performance in highly complex scenarios, highlighting its adaptability and utility in advancing context-aware drug relational learning.

Accurately predicting protein-ligand binding affinity is a critical challenge in drug discovery, crucial for understanding drug efficacy. While existing models typically rely on atom-level interactions, they often fail to capture the complex, higher-order interactions, resulting in noise and computational inefficiency. Transitioning to modeling these interactions at the cluster level is challenging because it is difficult to determine which atoms form meaningful clusters that drive the protein-ligand interactions. To address this, we propose CheapNet, a novel interaction-based model that integrates atom-level representations with hierarchical cluster-level interactions through a cross-attention mechanism. By employing differentiable pooling of atom-level embeddings, CheapNet efficiently captures essential higher-order molecular representations crucial for accurate binding predictions. Extensive evaluations demonstrate that CheapNet not only achieves state-of-the-art performance across multiple binding affinity prediction tasks but also maintains prediction accuracy with reasonable computational efficiency. The code of CheapNet is available at https://github.com/hyukjunlim/CheapNet.

The advent of generative AI now enables large-scale $\textit{de novo}$ design of molecules, but identifying viable drug candidates among them remains an open problem. Existing drug-likeness prediction methods often rely on ambiguous negative sets or purely structural features, limiting their ability to accurately classify drugs from non-drugs. In this work, we introduce BounDr.E: a novel modeling of drug-likeness as a compact space surrounding approved drugs through a dynamic one-class boundary approach. Specifically, we enrich the chemical space through biomedical knowledge alignment, and then iteratively tighten the drug-like boundary by pushing non-drug-like compounds outside via an Expectation-Maximization (EM)-like process. Empirically, BounDr.E achieves 10% F1-score improvement over the previous state-of-the-art and demonstrates robust cross-dataset performance, including zero-shot toxic compound filtering. Additionally, we showcase its effectiveness through comprehensive case studies in large-scale $\textit{in silico}$ screening. Our codes and constructed benchmark data under various schemes are provided at: https://github.com/eugenebang/boundr_e.

Accurate survival prediction is essential in healthcare as it guides treatment strategies and improves patient outcomes. While clinical features provide valuable prognostic information, they often fail to represent the molecular complexity of diseases. Transcriptomic data, which reflects gene expression patterns of tumors, present a complementary perspective to address this limitation. We introduce Transcriptome Transformer (TxT), a multitask learning framework that uses a transcriptome-centric approach to improve patient survival prediction. TxT employs a Transformer-based architecture with multihead attention mechanisms to effectively capture complex dependencies among genes, enabling dynamic modeling of gene-gene interactions while using shared information across multiple clinical prediction tasks. By jointly analyzing transcriptomic data and incorporating clinical features, TxT offers a more complete representation of patient biology. In experiments across both single-task and multitask datasets, TxT outperformed existing methods in survival prediction and related clinical tasks. Additionally, TxT offers biological insights through attention-derived gene interaction networks, identifying immune-related pathways in longer-surviving Luminal A patients and coagulation and epithelial-mesenchymal transition pathways in shorter-surviving counterparts. Differential attention analysis further revealed that integrating clinical features enhances the model's ability to prioritize genes involved in biologically meaningful pathways that are known to influence tumor progression and distant recurrence. The source code of TxT is available at https://github.com/BonilKoo/TxT.

Knowledge of unintended effects of drugs is critical in assessing the risk of treatment and in drug repurposing. Although numerous existing studies predict drug-side effect presence, only four of them predict the frequency of the side effects. Unfortunately, current prediction methods 1) do not utilize drug targets, 2) do not predict well for unseen drugs, and 3) do not use multiple heterogeneous drug features. We propose a novel deep learning-based drug-side effect frequency prediction model. Our model utilized heterogeneous features such as target protein information as well as molecular graph, fingerprints, and chemical similarity to create drug embeddings simultaneously. Furthermore, the model represents drugs and side effects into a common vector space, learning the dual representation vectors of drugs and side effects, respectively. We also extended the predictive power of our model to compensate for the drugs without clear target proteins using the Adaboost method. We achieved state-of-the-art performance over the existing methods in predicting side effect frequencies, especially for unseen drugs. Ablation studies show that our model effectively combines and utilizes heterogeneous features of drugs. Moreover, we observed that, when the target information given, drugs with explicit targets resulted in better prediction than the drugs without explicit targets.

Out-of-distribution (OOD) generalization in the graph domain is challenging due to complex distribution shifts and a lack of environmental contexts. Recent methods attempt to enhance graph OOD generalization by generating flat environments. However, such flat environments come with inherent limitations to capture more complex data distributions. Considering the DrugOOD dataset, which contains diverse training environments (e.g., scaffold, size, etc.), flat contexts cannot sufficiently address its high heterogeneity. Thus, a new challenge is posed to generate more seman-tically enriched environments to enhance graph invariant learning for handling distribution shifts. In this paper, we propose a novel approach to generate hierarchical seman-tic environments for each graph. Firstly, given an input graph, we explicitly extract variant subgraphs from the in-put graph to generate proxy predictions on local environ-ments. Then, stochastic attention mechanisms are employed to re-extract the subgraphs for regenerating global environ-ments in a hierarchical manner. In addition, we introduce a new learning objective that guides our model to learn the diversity of environments within the same hierarchy while maintaining consistency across different hierarchies. This approach enables our model to consider the relationships between environments and facilitates robust graph invariant learning. Extensive experiments on real-world graph data have demonstrated the effectiveness of our framework. Par-ticularly, in the challenging dataset DrugOOD, our method achieves up to 1.29% and 2.83% improvement over the best baselines on IC50 and EC50 prediction tasks, respectively.

Combination therapies have brought significant advancements to the treatment of various diseases in the medical field. However, searching for effective drug combinations remains a major challenge due to the vast number of possible combinations. Biomedical knowledge graph (KG)-based methods have shown potential in predicting effective combinations for wide spectrum of diseases, but the lack of credible negative samples has limited the prediction performance of machine learning models. To address this issue, we propose a novel model-agnostic framework that leverages existing drug-drug interaction (DDI) data as a reliable negative dataset and employs supervised contrastive learning (SCL) to transform drug embedding vectors to be more suitable for drug combination prediction. We conducted extensive experiments using various network embedding algorithms, including random walk and graph neural networks, on a biomedical KG. Our framework significantly improved performance metrics compared to the baseline framework. We also provide embedding space visualizations and case studies that demonstrate the effectiveness of our approach. This work highlights the potential of using DDI data and SCL in finding tighter decision boundaries for predicting effective drug combinations.

Drug response prediction (DRP) is important for precision medicine to predict how a patient would react to a drug before administration. Existing studies take the cell line transcriptome data, and the chemical structure of drugs as input and predict drug response as IC50 or AUC values. Intuitively, use of drug target interaction (DTI) information can be useful for DRP. However, use of DTI is difficult because existing drug response database such as CCLE and GDSC do not have information about transcriptome after drug treatment. Although transcriptome after drug treatment is not available, if we can compute the perturbation effects by the pharmacologic modulation of target gene, we can utilize the DTI information in CCLE and GDSC. In this study, we proposed a framework that can improve existing deep learning-based DRP models by effectively utilizing drug target information. Our framework includes NetGP, a module to compute gene perturbation scores by the network propagation technique on a network. NetGP produces genes in a ranked list in terms of gene perturbation scores and the ranked genes are input to a multi-layer perceptron to generate a fixed dimension vector for the integration with existing DRP models. This integration is done in a model-agnostic way so that any existing DRP tool can be incorporated. As a result, our framework boosts the performance of existing DRP models, in 64 of 72 comparisons. The performance gains are larger especially for test scenarios with samples with unseen drugs by large margins up to 34% in Pearson's correlation coefficient.

Recently, graph neural networks (GNNs) have been widely used for document classification. However, most existing methods are based on static word co-occurrence graphs without sentence-level information, which poses three challenges:(1) word ambiguity, (2) word synonymity, and (3) dynamic contextual dependency. To address these challenges, we propose a novel GNN-based sparse structure learning model for inductive document classification. Specifically, a document-level graph is initially generated by a disjoint union of sentence-level word co-occurrence graphs. Our model collects a set of trainable edges connecting disjoint words between sentences, and employs structure learning to sparsely select edges with dynamic contextual dependencies. Graphs with sparse structure can jointly exploit local and global contextual information in documents through GNNs. For inductive learning, the refined document graph is further fed into a general readout function for graph-level classification and optimization in an end-to-end manner. Extensive experiments on several real-world datasets demonstrate that the proposed model outperforms most state-of-the-art results, and reveal the necessity to learn sparse structures for each document.

BACKGROUND: The widely spreading coronavirus disease (COVID-19) has three major spreading properties: pathogenic mutations, spatial, and temporal propagation patterns. We know the spread of the virus geographically and temporally in terms of statistics, i.e., the number of patients. However, we are yet to understand the spread at the level of individual patients. As of March 2021, COVID-19 is wide-spread all over the world with new genetic variants. One important question is to track the early spreading patterns of COVID-19 until the virus has got spread all over the world. RESULTS: In this work, we proposed AutoCoV, a deep learning method with multiple loss object, that can track the early spread of COVID-19 in terms of spatial and temporal patterns until the disease is fully spread over the world in July 2020. Performances in learning spatial or temporal patterns were measured with two clustering measures and one classification measure. For annotated SARS-CoV-2 sequences from the National Center for Biotechnology Information (NCBI), AutoCoV outperformed seven baseline methods in our experiments for learning either spatial or temporal patterns. For spatial patterns, AutoCoV had at least 1.7-fold higher clustering performances and an F1 score of 88.1%. For temporal patterns, AutoCoV had at least 1.6-fold higher clustering performances and an F1 score of 76.1%. Furthermore, AutoCoV demonstrated the robustness of the embedding space with an independent dataset, Global Initiative for Sharing All Influenza Data (GISAID). CONCLUSIONS: In summary, AutoCoV learns geographic and temporal spreading patterns successfully in experiments on NCBI and GISAID datasets and is the first of its kind that learns virus spreading patterns from the genome sequences, to the best of our knowledge. We expect that this type of embedding method will be helpful in characterizing fast-evolving pandemics.

MOTIVATION: Identifying differentially expressed genes (DEGs) in transcriptome data is a very important task. However, performances of existing DEG methods vary significantly for data sets measured in different conditions and no single statistical or machine learning model for DEG detection perform consistently well for data sets of different traits. In addition, setting a cutoff value for the significance of differential expressions is one of confounding factors to determine DEGs. RESULTS: We address these problems by developing an ensemble model that refines the heterogeneous and inconsistent results of the existing methods by taking accounts into network information such as network propagation and network property. DEG candidates that are predicted with weak evidence by the existing tools are re-classified by our proposed ensemble model for the transcriptome data. Tested on 10 RNA-seq datasets downloaded from gene expression omnibus (GEO), our method showed excellent performance of winning the first place in detecting ground truth (GT) genes in eight datasets and find almost all GT genes in six datasets. On the other hand, performances of all existing methods varied significantly for the 10 data sets. Because of the design principle, our method can accommodate any new DEG methods naturally. AVAILABILITY: The source code of our method is available at https://github.com/jihmoon/MLDEG.

Breast cancer is classified into five intrinsic subtypes, with differing treatment methods and prognoses. Therefore, accurate identification of subtypes from patient transcriptome data is essential. Many gene signatures, including PAM50, have been developed to classify breast cancer subtypes. However, existing gene selection methods do not utilize biological pathways. Gene signature selection using biological pathways can explain signature genes in terms of biological functions. Thus, we propose a probabilistic model for pathway-guided gene set selection using gene expression data. First, we defined gene and pathway factors based on gene expression and pathway activation levels, and calculated the posterior probability. Second, we adopted the prediction strength to guide gene set selection. Third, the gene set was selected using the posterior probability and prediction strength values. Finally, on evaluating the selected gene set, it was experimentally confirmed that our gene set performed better on classification tasks than the PAM50 gene set, a gene set produced by the XGBoost classifier, and a random gene set. Among the genes selected by our method, it was confirmed that the genes included in the cell cycle and circadian rhythm pathways showed different expression patterns for each breast cancer subtype. Our selected gene set exhibited biological significance in terms of pathway activation.

MOTIVATION: Existing k-mer based string kernel methods have been successfully used for sequence comparison. However, existing kernel methods have limitations for comparative and evolutionary comparisons of genomes due to the sensitiveness to over-represented k-mers and variable sequence lengths. RESULTS: In this study, we propose a novel ranked k-spectrum string (RKSS) kernel. 1) RKSS kernel utilizes common k-mer sets across species, named landmarks, that can be used for comparing multiple genomes. 2) Based on the landmarks, we can use ranks of k-mers, rather than frequencies, that can produce more robust distances between genomes. To show the power of RKSS kernel, we conducted two experiments using 10 mammalian species with exon, intron, and CpG island sequences. RKSS kernel reconstructed more consistent evolutionary trees than the k-spectrum string kernel. In the subsequent experiment, for each sequence, kernel distance was calculated from 30 landmarks representing exon, intron, and CpG island sequences of 10 genomes. Based on kernel distances, concordance tests were performed and the result suggested that more information is conserved in CpG islands across species than in introns. In conclusion, our analysis suggests that the relational order, exon CpG island intron, in terms of evolutionary information contents.

MOTIVATION: Biological pathways are extensively used for the analysis of transcriptome data to characterize biological mechanisms underlying various phenotypes. There are a number of computational tools that summarize transcriptome data at the pathway level. However, there is no comparative study on how well these tools produce useful information at the cohort level, enabling comparison of many samples or patients. RESULTS: In this study, we systematically compared and evaluated 13 different pathway activity inference tools based on 5 comparison criteria using pan-cancer data set. This study has two major contributions. First, our study provides a comprehensive survey on computational techniques used by existing pathway activity inference tools. The tools use different strategies and assume different requirements on data: input transformation, use of labels, necessity of cohort-level input data, use of gene relations and scoring metric. Second, we performed extensive evaluations on the performance of these tools. Because different tools use different methods to map samples to the pathway dimension, the tools are evaluated at the pathway level using five comparison criteria. Starting from measuring how well a tool maintains the characteristics of original gene expression values, robustness was also investigated by adding noise into gene expression data. Classification tasks on three clinical variables (tumor versus normal, survival and cancer subtypes) were performed to evaluate the utility of tools for their clinical applications. In addition, the inferred activity values were compared between the tools to see how similar they are along with the scoring schemes they use.

MOTIVATION: Biological pathway is an important curated knowledge of biological processes. Thus, cancer subtype classification based on pathways will be very useful to understand differences in biological mechanisms among cancer subtypes. However, pathways include only a fraction of the entire gene set, only one-third of human genes in KEGG, and pathways are fragmented. For this reason, there are few computational methods to use pathways for cancer subtype classification. RESULTS: We present an explainable deep-learning model with attention mechanism and network propagation for cancer subtype classification. Each pathway is modeled by a graph convolutional network. Then, a multi-attention-based ensemble model combines several hundreds of pathways in an explainable manner. Lastly, network propagation on pathway-gene network explains why gene expression profiles in subtypes are different. In experiments with five TCGA cancer datasets, our method achieved very good classification accuracies and, additionally, identified subtype-specific pathways and biological functions. AVAILABILITY AND IMPLEMENTATION: The source code is available at http://biohealth.snu.ac.kr/software/GCN_MAE. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Characterizing cancer subclones is crucial for the ultimate conquest of cancer. Thus, a number of bioinformatic tools have been developed to infer heterogeneous tumor populations based on genomic signatures such as mutations and copy number variations. Despite accumulating evidence for the significance of global DNA methylation reprogramming in certain cancer types including myeloid malignancies, none of the bioinformatic tools are designed to exploit subclonally reprogrammed methylation patterns to reveal constituent populations of a tumor. In accordance with the notion of global methylation reprogramming, our preliminary observations on acute myeloid leukemia (AML) samples implied the existence of subclonally occurring focal methylation aberrance throughout the genome. RESULTS: We present PRISM, a tool for inferring the composition of epigenetically distinct subclones of a tumor solely from methylation patterns obtained by reduced representation bisulfite sequencing. PRISM adopts DNA methyltransferase 1-like hidden Markov model-based in silico proofreading for the correction of erroneous methylation patterns. With error-corrected methylation patterns, PRISM focuses on a short individual genomic region harboring dichotomous patterns that can be split into fully methylated and unmethylated patterns. Frequencies of such two patterns form a sufficient statistic for subclonal abundance. A set of statistics collected from each genomic region is modeled with a beta-binomial mixture. Fitting the mixture with expectation-maximization algorithm finally provides inferred composition of subclones. Applying PRISM for two AML samples, we demonstrate that PRISM could infer the evolutionary history of malignant samples from an epigenetic point of view. AVAILABILITY AND IMPLEMENTATION: PRISM is freely available on GitHub (https://github.com/dohlee/prism). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: The main research topic in this paper is how to compare multiple biological experiments using transcriptome data, where each experiment is measured and designed to compare control and treated samples. Comparison of multiple biological experiments is usually performed in terms of the number of DEGs in an arbitrary combination of biological experiments. This process is usually facilitated with Venn diagram but there are several issues when Venn diagram is used to compare and analyze multiple experiments in terms of DEGs. First, current Venn diagram tools do not provide systematic analysis to prioritize genes. Because that current tools generally do not fully focus to prioritize genes, genes that are located in the segments in the Venn diagram (especially, intersection) is usually difficult to rank. Second, elucidating the phenotypic difference only with the lists of DEGs and expression values is challenging when the experimental designs have the combination of treatments. Experiment designs that aim to find the synergistic effect of the combination of treatments are very difficult to find without an informative system. RESULTS: We introduce Venn-diaNet, a Venn diagram based analysis framework that uses network propagation upon protein-protein interaction network to prioritizes genes from experiments that have multiple DEG lists. We suggest that the two issues can be effectively handled by ranking or prioritizing genes with segments of a Venn diagram. The user can easily compare multiple DEG lists with gene rankings, which is easy to understand and also can be coupled with additional analysis for their purposes. Our system provides a web-based interface to select seed genes in any of areas in a Venn diagram and then perform network propagation analysis to measure the influence of the selected seed genes in terms of ranked list of DEGs. CONCLUSIONS: We suggest that our system can logically guide to select seed genes without additional prior knowledge that makes us free from the seed selection of network propagation issues. We showed that Venn-diaNet can reproduce the research findings reported in the original papers that have experiments that compare two, three and eight experiments. Venn-diaNet is freely available at: http://biohealth.snu.ac.kr/software/venndianet.