Discover, rate, and discuss biology preprints.

This study benchmarks agentic large language models for protein-set functional annotation, finding moderate reproducibility and common error modes, but improved evidence handling and reproducibility.

This paper introduces a computational pipeline combining image processing and machine learning to segment and quantify kidney tissue structures using Normalized Raman Imaging, achieving high segmentation performance.

This study introduces ECIFP, a method to analyze ligand binding modes in the PDB, showing it can identify similar binding sites and detect binding-mode similarities across diverse ligands.

This paper benchmarks spatial transcriptomics in stem cell-derived organoids using Stereo-seq, optimizing RNA capture and introducing regional analysis methods.

This study evaluates segmentation models on live microscopy and spatial-omics data, finding that no single model consistently outperforms others across all datasets.

This paper introduces trRosettaRNA2, a deep learning model for RNA 3D structure prediction that outperforms existing methods with fewer parameters, and demonstrates its effectiveness on RNase P RNA.

The paper develops a structure-informed Siamese GNN to classify CirA missense variants affecting cefiderocol susceptibility using synthetic data.

SpaFlow models ligand-receptor interactions in spatial transcriptomics data by simulating diffusion and binding processes, revealing spatially coherent signaling patterns in multiple disease contexts.

PathPinpointR predicts scRNA-seq sample positions along reference trajectories using switching-gene events to assign pseudo-time values.

MUSTARD reduces dimensionality of multi-sample scRNA-seq data using trajectory guidance, revealing sample-level heterogeneity and associated gene markers.

CSsingle improves bulk and spatial transcriptomic data decomposition by integrating with scRNA-seq, showing better accuracy and revealing MCCs as potential biomarkers in EAC.

The paper introduces MuTopia, a framework that identifies topographic archetypes of mutagenesis in cancer genomes, revealing that mutational processes are constrained by replication timing and chromatin state.

GNOMES normalizes and analyzes ChIP-seq and CUT&RUN data, providing differential binding results and quality control metrics through an integrated workflow.

The paper introduces diagFDR, a framework to improve FDR reporting in proteomics by assessing scope, calibration, and stability, and provides tools for verification.

The paper introduces UKBioBERT, a DNA language model pre-trained on UK BioBank variants, which improves gene expression prediction and enables eQTL identification when combined with sequence-to-function models.

This study compares co-folding and docking methods for drug design, finding that physics-based docking outperforms co-folding for novel ligands and provides physical rationale for binding.

This study benchmarks five deep learning models for predicting Hi-C contact maps, finding Epiphany performs best in accuracy, generalization, and loop detection.

This study compares four proteomics software tools for hair proteome analysis, finding differences in peptide identification, protein inference, functional annotation, and quantification across tools and extraction methods.

The paper develops embeRNA, a method to predict microbial functions from 16S rRNA sequences without reference databases, showing better performance in novel environments.

This study evaluates the predictive power of genomic and clinical features in pan-cancer survival models, finding that clinical variables dominate over genomic features.