Discover, rate, and discuss biology preprints.

The paper evaluates peptide-MHC generalization to unseen alleles and introduces a benchmark and model that improves prediction accuracy for rare HLA alleles.

HoloCell is a generative foundation model that integrates epigenomics, transcriptomics, and proteomics data across single-cell multi-omics modalities, demonstrating superior performance in representation learning and cross-modal generation.

This paper compares output-head designs for predicting diabetes and myocardial infarction complications, finding that symmetric CRF outperformed others in diabetes but no single method dominated in myocardial infarction.

FFixR filters artefactual mutations in FFPE RNA-seq data, improving accuracy for somatic variant calling in cancer.

This study uses computational methods to analyze the binding of three withanolide derivatives to Plasmodium falciparum DNA gyrase, showing their potential as antimalarial drug candidates.

FateLimit quantifies when future cell identities become predictable from current molecular states across multiple developmental systems.

OmniCell integrates single-cell and spatial transcriptomics to model tissue context, revealing how gene programs operate in different anatomical environments.

biomeStat automates genomic epidemiology analysis of 1,000 Dengue virus genomes, revealing immune escape sites and conserved drug targets in under 24 hours.

This paper develops a deep learning framework to predict protein flexibility from static structural data, achieving moderate accuracy and identifying key structural features associated with flexibility.

VCBench evaluates single-cell foundation models across seven dimensions, finding baselines often outperform models, with TranscriptFormer showing unique strengths but tradeoffs.

The paper uses metagenomic analysis to identify ancient organisms in a corn mummy from ancient Egypt, finding potential historical contaminants alongside modern ones.

The paper develops a multi-model consensus deep learning approach for automated gold fiducial marker segmentation in prostate MRI, achieving high sensitivity and precision.

This paper develops a multi-scale machine learning framework to predict antibody-antigen binding affinity using deep mutational scanning and structural features, achieving improved performance over single-modality approaches.

This study develops seqLens, a genomic language model family, and evaluates its performance on phenotypic predictions using diverse pretraining datasets and tokenization strategies.

The paper introduces MetagenomicKG, a knowledge graph integrating metagenomic data with biomedical knowledge to enable hypothesis generation and pathogen prediction.

The paper extends the ColBuilder framework to include non-enzymatic AGE crosslinks in collagen fibril simulations, showing their distinct mechanical effects compared to enzymatic crosslinks.

DiaReport provides a reproducible workflow for differential expression analysis and interactive reporting in DIA-based proteomics, demonstrating its utility with an extracellular vesicle dataset.

EMAlign improves cryo-EM map alignment using deep learning to predict main-chain probability, achieving high accuracy and success rates across global and local alignments.

PISTACHIO improves spatial cell-type deconvolution in tumours using proteomics constraints and non-negative matrix factorization.

This paper introduces a random forest-based method for multi-omics integration that separates shared and modality-specific signals, showing improved clustering and biological insights in simulations and real datasets.