Briefings in bioinformatics最新文献

Deep learning and machine learning models have shown promise in drug response prediction (DRP), yet their ability to generalize across datasets remains an open question, raising concerns about their real-world applicability. Due to the lack of standardized benchmarking approaches, model evaluations and comparisons often rely on inconsistent datasets and evaluation criteria, making it difficult to assess true predictive capabilities. In this work, we introduce a benchmarking framework for evaluating cross-dataset prediction generalization in DRP models. Our framework incorporates five publicly available drug screening datasets, seven standardized DRP models, and a scalable workflow for systematic evaluation. To assess model generalization, we introduce a set of evaluation metrics that quantify both absolute performance (e.g. predictive accuracy across datasets) and relative performance (e.g. performance drop compared to within-dataset results), enabling a more comprehensive assessment of model transferability. Our results reveal substantial performance drops when models are tested on unseen datasets, underscoring the importance of rigorous generalization assessments. While several models demonstrate relatively strong cross-dataset generalization, no single model consistently outperforms across all datasets. Furthermore, we identify CTRPv2 as the most effective source dataset for training, yielding higher generalization scores across target datasets. By sharing this standardized evaluation framework with the community, our study aims to establish a rigorous foundation for model comparison, and accelerate the development of robust DRP models for real-world applications.

Alternative polyadenylation (APA) of $3^{prime}$untranslated regions ($3^{prime}$UTRs) is a pervasive mechanism that regulates mRNA stability, localization, and translational efficiency by generating isoforms with distinct $3^{prime}$UTR lengths and regulatory element composition. Despite its critical role in fine-tuning gene expression, APA has been largely overlooked in transcriptome-wide association studies (TWAS), which traditionally rely on linear models of SNP effects. To bridge this gap, we developed ASTWAS, a two-stage framework that first trains APA usage prediction models (BLUP, Elastic Net, LASSO, and TOP1) to quantify SNP impacts on distal poly(A) site choice via the percentage of distal poly(A) site usage index, and then aggregates weighted SNP effects within a kernel method to capture both linear and nonlinear genetic interactions. In extensive simulations spanning additive, epistatic, heterogeneous, compensatory, and single-variant architectures under both pleiotropy and causality scenarios, ASTWAS shows higher statistical power than linear APA-TWAS ($3^{prime}$aTWAS), especially at low heritability and in the presence of SNP interactions. Applied to WTCCC type 1 diabetes and rheumatoid arthritis cohorts, ASTWAS not only rediscovers known susceptibility genes but also suggests novel candidates (e.g. GABBR1, RGL2) that form coherent interaction modules and enrich immune-related pathways, underscoring the biological significance of our algorithm in complex trait genetics. ASTWAS is implemented in Python and freely available at https://github.com/wl-Simplecss/ASTWAS.

Protein language models (pLMs) have become essential tools in computational biology, powering diverse applications from variant effect prediction to protein engineering. Central to their success is the use of pretrained embeddings-contextualized representations of amino acid sequences-which enable effective transfer learning, especially in data-scarce settings. However, recent studies have revealed that standard masked language modeling objectives used to train these models often produce representations that are misaligned with the needs of downstream tasks. While scaling up model size improves performance in some cases, it does not universally yield better representations. In this study, we investigate two complementary strategies for improving pLM representations: (i) integrating text annotations through contrastive learning, and (ii) combining multiple embeddings via embedding fusion. We benchmark six text-integrated pLMs (tpLMs) and three large-scale pLMs across six biologically diverse tasks, showing that no single model dominates across settings. Fusion of multiple tpLMs embeddings improves performance on most tasks but presents a computational bottleneck due to the combinatorial number of possible combinations. To overcome this, we propose greedier forward selection, a linear-time algorithm that efficiently identifies near-optimal embedding subsets. We validate its utility through two case studies, homologous sequence recovery and protein-protein interaction prediction, demonstrating new state-of-the-art results in both. Our work highlights embedding fusion as a practical and scalable strategy for improving protein representations.

Cancer development is driven by somatic evolution and clonal selection. However, traditional selective pressure analysis methods have treated all sites within a gene equally, such a gene-level model oversimplifies the complexity of cancer evolution. In this study, we introduced CN/CS-calculator, a novel site-specific method that can capture selective pressures acting across different gene sites. By deciphering the interplay between the selection pattern and the function of a gene in oncogenesis, CN/CS-calculator uncovers a unique class of mini-driver genes, which exhibit weak positive selection, with certain critical sites providing context-dependent promoter effects on the fitness of cancer subclones while others are constrained by evolutionary conservation. Our method emphasizes the importance of site-specific analysis in uncovering how subtle evolutionary forces shape cancer biology. The refined understanding offers new insights into the mechanisms of cancer heterogeneity and molecular evolution, with potential implications for advancing therapeutic strategies and prognostic assessments.

Identifying transcription factors (TFs) responsible for gene expression changes remain a central challenge in functional genomics. TFEA.ChIP is a ChIP-seq-based TF enrichment analysis tool that addresses this by linking TF binding profiles to differentially expressed genes through experimentally supported cis-regulatory element (CRE)-gene associations. Unlike motif- or heuristic-based approaches, TFEA.ChIP adopts a biologically grounded strategy by intersecting TF binding data from ReMap2022 with regulatory maps from ENCODE's rE2G and CREdb. To overcome the high context-specificity of rE2G associations, we developed filtering strategies based on confidence scores and recurrence across biosamples. Benchmarking on 342 curated gene sets from the Molecular Signatures Database C2 CGP collection showed that recurrence-based filtering significantly improved accuracy, outperforming the original GeneHancer-based implementation and leading tools including BARTv2.0, Lisa, ChEA3, and HOMER. A case study on hypoxia further validated the method, demonstrating accurate and pathway-specific enrichment of hypoxia-inducible factor-related TFs using both overrepresentation analysis and gene set enrichment analysis. Additionally, the updated implementation of TFEA.ChIP in R/Bioconductor introduces several user-friendly features, including automated analysis workflows and expression-based filtering of candidate TFs. These additions streamline the integration of TFEA.ChIP into standard RNA-seq analysis pipelines, enabling more efficient and reproducible workflows. Together with its strong benchmarking performance and biologically grounded framework, the updated tool provides a robust and accessible solution for inferring transcriptional regulators from gene expression data.

Motivation: Recent advances in single-cell sequencing have transformed precise measurement of gene expression at cellular resolution, enabling unprecedented dissection of cellular heterogeneity and intricate biological processes. The accumulation of multi-omics data offers new avenues for cell clustering-a critical foundation for cell-type identification and downstream analyses. However, substantial challenges persist in simultaneously achieving effective integration of complementary information in multi-omics data and their appropriate weight allocation.

Results: Here, we propose an Adaptive Multi-View clustering framework with the Information Bottleneck principle to solve the multi-omics data clustering task (named scAMVIB). The proposed model could learn multi-view omics representations that capture both inter-omics associations and omics-specific patterns, with the adaptive weight allocation. Specifically, multi-view data comprise two components: (i) the integrated omics feature matrix derived from the similarity network fusion strategy and (ii) omics-specific representations from distinct platforms. These inputs are processed through a multi-view information bottleneck clustering framework that leverages cross-view complementarity to enhance representations. View weights are adaptively assigned via maximum entropy regularization, proportional to their information content. The final cell partitions are obtained through sequential iterative optimization. Comprehensive experiments across multiple datasets demonstrate that scAMVIB has strong competitiveness in clustering while maintaining biological interpretability.

RNA performs a variety of functions within cells and is implicated in various human diseases. Because druggable proteins occupy a small portion of the genome, considerable interest has been increasing in developing drugs targeting RNAs. Thus, precise prediction of small-molecule binding sites across different classes of RNAs is important. In this study, a lightweight deep learning program for predicting RNA-drug binding sites, called compound binding site prediction for RNA (CoBRA), is introduced. Our approach utilizes residue-level embeddings derived from a pre-trained RNA language model, without relying on any structural information. These embeddings encapsulate the contextual and statistical properties of each nucleotide and are used as input for a multi-layer perceptron classifier that performs binary classification of binding nucleotides. The model was trained using the TR60 and HARIBOSS datasets and tested on four independent benchmark sets. The performance of CoBRA demonstrates a relative improvement of 22.1% in the Matthew correlation coefficient and a 45.6% increase in sensitivity compared to existing state-of-the-art RNA-ligand binding site prediction methods that utilize structural information. These results demonstrate that sequence-based language model embeddings, which do not require explicit coordinate or distance information, can match or outperform structure-based methods. This makes it a flexible tool for predicting binding sites across diverse RNA targets.

The ability of T-cell receptors (TCRs) to recognize neoantigens is fundamental to the initiation and maintenance of adaptive immune responses. In TCR-based immunotherapies, elucidating the recognition patterns of TCRs for peptides and accurately identifying therapeutically relevant TCR-peptide pairs remain critical challenges. Here, we present a novel dual-pathway network model, ProTCR, which integrates the protein language model ProtT5 with deep learning methods. By incorporating both global and local feature extraction mechanisms, ProTCR enables efficient representation of amino acid sequences, thereby enhancing the model's generalizability across diverse data distributions and improving its biological interpretability. ProTCR demonstrates robust performance and broad applicability across various datasets, including neoantigens, previously unseen peptides, and MHC class II-restricted epitopes, overcoming the reliance on known peptide-TCR pairs observed in previous studies. It also offers new insights for predicting diverse classes of antigenic peptides. We applied ProTCR to several clinically relevant scenarios, including immunotherapeutic target identification in acute myeloid leukemia, neoantigen-targeted immunotherapy in solid tumours, and antigen-specific T cell recognition against pathogens such as influenza and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Across these complex settings, ProTCR consistently maintained high accuracy and stability, demonstrating strong cross-task adaptability and broad potential for clinical application. This work not only provides a powerful tool for elucidating immune response mechanisms but also offers a solid computational foundation for the design of neoantigen or TCR based precision immunotherapy strategies.

Enzyme catalysis, with its advantages in environmental sustainability and efficiency, is gaining traction across diverse industrial applications, such as waste utilization and pharmaceutical biomanufacturing. However, optimizing enzyme catalytic activity remains a significant challenge. To facilitate enzyme mining and engineering, machine learning (ML) models have emerged to predict enzyme substrate specificity, enzyme turnover number, and enzyme catalytic optimum. This review endeavored to assist researchers in effectively utilizing predictive models for enzyme catalytic activity through presenting recent advancements and analyzing different approaches. We also pointed out existing limitations (e.g. dataset imbalance) and offered suggestions on potential enhancements to address them. We identified that the attention mechanism, inclusion of new features such as product information and temperature, and using transfer learning to leverage different datasets were three main useful modeling strategies. Furthermore, we envisaged that accurate predictors of enzyme catalytic activity would potentially transform enzyme and metabolic engineering, and the optimization of biocatalysis.