Discover, rate, and discuss biology preprints.

This paper develops a tensor decomposition framework to integrate single-cell four-omics data, revealing cell-type-specific enhancer-promoter programs that cannot be identified by individual modalities.

LifeSciBench evaluates language models on 750 expert-authored tasks across seven life science domains and workflows, revealing significant gaps in model performance.

The paper compares sequence-derived representations and Pfam-domain content for retrieving biosynthetic gene clusters, finding that Pfam remains the dominant signal.

The paper benchmarks 16S amplicon sequencing tools and databases using mock communities to evaluate their accuracy in estimating microbial diversity.

This study benchmarks generative AI tools for summarizing genomic variant literature, finding VarChat and Gpt-4o perform best with expert validation needed for reliability.

HNSW-MS enables fast and accurate spectral search in large metabolomics repositories, allowing rapid integration of new spectra into global molecular networks.

OncoGenRAG improves variant interpretation in oncology by combining retrieval and classification with a curated knowledge base, achieving high accuracy and reducing hallucinations.

This paper describes a persistent AI scientist system that operates continuously, improves through memory and feedback, and generates hypotheses and scientific outputs over time.

The paper develops a server enabling large language models to access drug-gene interaction data through natural language queries.

This study evaluates scRNA-seq foundation models for plant perturbation response prediction, finding that performance varies significantly with evaluation protocol and that preserving gene identity improves generalization.

The paper introduces disize, a normalization method that uses experimental design information to improve size factor estimation in RNA-seq data, showing better accuracy in challenging scenarios.

This study harmonizes host and vector omics data to uncover coordinated molecular interactions during Begomovirus transmission, identifying key regulatory modules and targets for RNA interference strategies.

AlphaConformers improves AlphaFold2 by generating multiple protein conformations using structural information from databases, outperforming existing methods in capturing alternative states.

The paper introduces MetaChlam, a pipeline for strain typing Chlamydia trachomatis in metagenomes, showing it outperforms existing tools and detects contaminants.

The paper presents a framework integrating AI and computational tools to evaluate CAR target characteristics, using HER2 as a case study to identify liabilities beyond conventional expression analysis.

This paper introduces a benchmark dataset for enzyme function prediction and shows that current models struggle to distinguish non-functional enzyme variants from functional ones.

This paper introduces Dynamics-Aware Evolutionary Profiling to separate structural rigidity from functional motion in proteins, improving variant interpretation by identifying dynamic-conserved residues.

CLEAR-ST corrects mRNA lateral diffusion in spatial transcriptomics by modeling contamination with a physics-informed probabilistic framework, improving spatial domain recovery and downstream analysis accuracy.

This paper introduces EvoBind-multimer, a deep learning framework for designing molecular glues from protein sequences, demonstrating its ability to recruit E3 ligases to oncoproteins and showing context-dependent functionality.

The paper maps pathogenicity patterns in GLUT transporters by analyzing missense mutations and their structural impact across the family.