Discover, rate, and discuss biology preprints.

This study benchmarks 86 heritability estimation configurations and evaluates their impact on polygenic risk score performance across 10 UK Biobank phenotypes.

This study compares machine learning models for SARS-CoV-2 variant classification under imbalanced data, finding classical models with TF-IDF k-mer features outperform deep learning methods.

Cenote-Taker 3 is a tool for virus discovery and annotation in genomic data, outperforming existing tools in speed and accuracy.

CLEAR improves gene set enrichment analysis by jointly modeling gene sets with continuous statistics, reducing redundancy and improving sensitivity.

The paper shows that Wnt signaling in mesenchymal TNBC cells increases inflammation and EMT-related gene expression, and identifies a 55-gene signature associated with these effects.

GFETM improves scATAC-seq modeling by integrating genome foundation model embeddings, revealing cell-state specific TF activity and kidney diabetes epigenomic signatures.

HalluCodon designs species-specific coding sequences using multimodal language models trained on plant omics data, generating codons that match host-specific usage and support high expression.

mRNA-GPT is a generative model that optimizes full-length mRNA sequences by jointly optimizing all three regions, achieving Pareto-optimal designs and outperforming existing methods.

The paper introduces GPCT, a pretrained transformer model for cytometry that improves sample-level prediction and interpretability across diverse datasets.

This study determines the minimum sleep data duration needed for reliable circadian phase estimation and develops new methods to reduce this requirement.

This paper benchmarks three agentic bioinformatics systems for retrieving proteins related to coccolithophore calcification, finding that Codex best balances sensitivity and specificity while DeerFlow and Biomni produce larger but less relevant sets.

The paper identifies methodological errors in epitope discovery that inflate in silico benchmarks and introduces a model trained on clean data that outperforms existing methods.

GeoBridge transforms sparse scRNA-seq data into a continuous landscape for reconstructing and navigating cellular state transitions.

This paper introduces scRitmo, a method to infer circadian phases and quantify biological desynchrony from single-cell transcriptomes, showing improved performance over existing methods.

The paper benchmarks protein language models for EC number prediction, finding MLP classifiers outperform BLAST and PLMs show significant gains for distant eukaryotes.

This paper develops a dynamic multimodal model for survival prediction in multiple myeloma using gene expression, lab data, and treatment history, achieving higher accuracy than existing methods.

The paper develops a machine learning model to predict protein detectability from RNA expression and subcellular localization across cancers, finding significant improvements when incorporating localization data.

This study benchmarks somatic mutation detection using the human pangenome reference and finds improved accuracy, especially in East Asian individuals.

STAPLE automates spatial transcriptomics analysis by integrating multiple tools into a modular framework and provides AI-driven interpretation of results.

The paper introduces SPLASH, a reference-free algorithm for detecting sequence variation from RNA-seq data, and applies it to discover novel regulatory mechanisms in plant transcriptomes.