The gut microbiome shapes cancer therapy response through its influence on host metabolism. While prior studies examine pairwise associations between individual genera and metabolites, there is limited methodology for identifying microbial genera that systematically regulate the overall metabolome. Scalable statistical tools are needed to uncover such system-level "master predictors" in high-dimensional microbiome-metabolome data.

To enable real-world protein-ligand affinity prediction, not only out-of-distribution generalization but also robustness to variable structural availability and quality should be considered in model design.

Drug repositioning accelerates clinical translation by identifying new therapeutic indications for approved drugs. However, therapeutic associations in biomolecular networks often exist indirectly, through transitive chains and long-range mechanisms, rather than as directly observed links. Shallow methods are confined to direct similarity and miss such indirect associations, whereas deep graph neural networks suffer from over-smoothing and lose discriminative power in highly connected networks.

Predicting spatial gene expression from histology images offers a cost-effective complement to spatial transcriptomics. However, existing methods struggle to balance spatial continuity with functional heterogeneity, often producing over-smoothed predictions or neglecting spatial context.

One of the challenges encountered when merging heterogeneous observational medical datasets is the recoding of categorical target variables that may have been measured differently across data sources. This study compares standard machine learning-based approaches, specifically Multiple Imputation by Chained Equations (MICE), k-Nearest Neighbours (kNN), missForest, and Factor Analysis of Mixed Data (missMDA), with an Optimal Transport-based algorithm (OTrecod). Their empirical performance is evaluated in realistic data integration settings that incorporate missing covariate values, non-linear relationships, and imbalanced groups, all of which remain underexplored.

Artificial intelligence accelerates anticancer peptides (ACPs) discovery. However, existing computational methods lack integration of identification with activity-based candidate prioritization. Here, we present DeepACPred, a three-stage pipeline encompassing ACP binary classification model, ACP multilabel classification model, and ACP IC50 prediction model, leveraging multimodal features from ESM2 protein language model embeddings, AAindex physicochemical descriptors, and sequence composition. On 5712 benchmark sequences, the binary classifier achieved 95.10% accuracy (AUC = 0.9913), with performance remaining stable under CD-HIT cluster-aware splitting at 40%-90% identity thresholds. Multilabel cancer-type prediction yielded macro-F1 = 0.9124 across seven cancer types, and log10(IC50) regression achieved Spearman ρ = 0.8602 under 5-fold cross-validation. Ablation experiments showed task-dependent feature contributions rather than uniformly additive multimodal effects. Applied to 260 000 motif-enriched 18-mer candidates, DeepACPred selected 12 peptides predicted to be active against breast cancer cells, all of which showed measurable in vitro cytotoxic activity against murine 4T1 cells in OD-derived dose-response assays (IC50: 0.88-36.83 μg/ml). Although prospective IC50 ranking showed limited fine-grained resolution, these results support the use of the regression module for coarse candidate enrichment. In conclusion, DeepACPred provides a systematic framework for ACP candidate enrichment and prioritization.

Accurate sequencing depth planning for UMI-based transcriptomic experiments currently relies on heuristic metrics such as reads per cell and sequencing saturation. However, sequencing saturation cannot uniquely determine molecular recovery because the relationship between these quantities depends on amplification heterogeneity.

Enrichment analysis is a cornerstone of "omics" data interpretation, enabling researchers to connect analysis results to biological processes and generate testable hypotheses. Enrichment analysis in metabolomics poses distinct challenges for interpretation and multi-omics integration due to the lack of well-defined and consistent connections to well-curated gene-centered biological knowledge repositories. To address these challenges, we developed hypeR-GEM, a methodology and associated R package that adapts gene set enrichment analysis to metabolomics. hypeR-GEM leverages genome-scale metabolic models (GEMs) to infer reaction-based links between metabolites and enzyme-coding genes, enabling the mapping of metabolite signatures to gene signatures and their subsequent annotation via gene set enrichment analysis.

Machine Learning Force Fields (MLFFs) have emerged as promising tools for accelerating molecular dynamics simulations. However, existing approaches often struggle to capture the geometric characteristics of long-range interactions, including distance and direction, remain sensitive to conformational variations, and lack adaptive mechanisms for balancing short- and long-range forces. To address these limitations, we propose GeoNet , a physicochemical-principle-guided framework for modeling dual-range atomic interactions. GeoNet employs geometric attention over atom-fragment bipartite graphs to characterize long-range dependencies, introduces dual-level augmentation to enforce semantic consistency across molecular conformations, and uses an adaptive fusion module to dynamically balance short- and long-range interaction pathways according to local atomic environments.

The rapid advancement of spatial omics is fundamentally hindered by the resolution gap between physical capture platforms and genuine biological microstructures, a challenge compounded by inherent data sparsity and noise. While current image-guided computational methods attempt to bridge this gap, they often lack the multi-modal flexibility, non-linear modeling, and scalability required for modern, whole-tissue datasets.

Whole-slide image (WSI)-based biomarker prediction in computational pathology has the potential to support scalable and resource-efficient analysis of large pathology cohorts, helping prioritize cases for downstream molecular testing and patient stratification. However, reliable biomarker labels are often limited and costly to obtain, making label-efficient WSI prediction essential. Recent slide-level foundation models have opened new opportunities for few-shot biomarker prediction by providing strong pretrained slide representations. Nevertheless, few-shot learning still suffers from sparse labeled support data, and data augmentation remains important for improving robustness and generalization. In frozen multi-stage WSI pipelines, however, conventional augmentation is difficult to apply: pixel-level augmentation requires costly feature re-extraction, while naive perturbation of frozen representations may compromise semantic consistency.

Fragment-based drug discovery (FBDD) efficiently explores chemical space by combining small molecular fragments. Advances in computational methods are accelerating the development of algorithm- and AI-based approaches in FBDD. However, it should be noted that certain methods do not provide synthetic pathways to obtain the proposed compounds. Consequently, these molecules might not be synthesized easily.

Long-read sequencing enables improved genome inference but remains challenged by high error rates that lead to excessive false-positive (FP) variant calls, particularly for small INDELs in complex genomic regions. We present VCboost, a deep learning-based post-calling framework designed to reduce FP in single-nucleotide polymorphism (SNP) and indel detection from long-read sequencing data. VCboost extracts discriminative features from pileup reads and consensus sequences and employs a dedicated filtering model integrating convolutional and recurrent neural networks with residual connections and multi-head attention. Evaluated on multiple human long-read datasets, VCboost consistently improved variant calling performance over Clair3, achieving a 5%-8% increase in precision and a 2%-5% gain in F1-score for INDELs, with minimal recall loss. For SNPs, VCboost improved precision by 4%-8% and F1-score by 2%-5%, with negligible impact on recall. Substantial performance gains were observed in difficult-to-map regions, including low-mappability and segmental-duplication regions, where SNP precision increased by up to 17.6%. Overall, VCboost effectively enhances the accuracy of long-read variant calling while preserving high sensitivity, offering a robust solution for variant detection in challenging genomic regions.

Mitochondrial RNA processing directed by the transfer ribonucleic acid (tRNA) punctuation model is essential for function and linked to human diseases. Strand-specific RNA sequencing can capture cleavage intermediates as reads with soft-clipping (unmapped sequences at read ends), but these signatures lack systematic characterization, limiting reliable cleavage site identification. We analyzed strand-specific RNA-seq data from 54 samples (35 private, 19 public) encompassing two library types. Soft-clipped reads were evaluated for frequency, quality, guanine-cytosine (GC) content, and fragment size, with sequence-level analysis of clipped portions. We compared random versus non-random priming across 10 sample pairs and assessed alignment strategies. Leveraging multiple features, we developed a random forest model to identify high-confidence cleavage sites and applied it to 20 hepatocellular carcinoma samples. Soft-clipping was prevalent in both library types but significantly higher in second-strand-specific libraries (P < 0.0001), independent of quality metrics. Soft-clipped sequences were predominantly 1-6 nt (87.9%-97.0%), guanine-rich, and preferentially at 3' ends (84.9%-93.8%). Random priming drove high-level 3' soft-clipping on both H-strand (54.47%) and L-strand (28.07%) transcripts, while non-random primers yielded minimal levels (<1.5%). Allowing soft-clipping during alignment increased sequencing depth and precision (P < 0.0001). The random forest model achieved excellent performance (F1 > 0.85, area under the curve > 0.90), with 1-2 nt soft-clips providing the highest signal-to-noise ratio. This first systematic characterization of soft-clipping in mitochondrial RNA-seq establishes a high-fidelity, machine-learning-based workflow for identifying cleavage sites, offering an accessible tool to advance studies of mitochondrial post-transcriptional regulation.