Discover, rate, and discuss biology preprints.

eRNAformer maps enhancer-derived RNA loci genome-wide using deep learning, identifying new eRNA loci linked to disease risk and cancer therapy.

The paper evaluates KAN variants and protein embeddings for enzyme classification, showing improved performance with ESM-2 embeddings.

G-LATO infers spatial gradients using deep Gaussian processes, revealing known and novel spatial organization in liver and glioblastoma data.

This paper investigates how pooling data from different biological populations affects gene co-expression network analysis and finds that it disrupts module structure and is undetectable by common metrics.

The paper shows that pooled correlations in transcriptomic data often reverse sign when analyzed within specific contexts, highlighting the need for context-aware reporting of correlations.

The paper reanalyzes a public single-cell RNA sequencing dataset to identify chondrocyte state transitions and ECM remodeling in osteoarthritis, finding collagen network hubs as potential targets.

MBIOME provides an open-source workflow for amplicon-based microbiome analysis, validated across sequencing platforms and showing improved flexibility and taxonomic resolution.

Prot2Prop predicts six protein developability properties using a multitask framework with a frozen ProstT5 encoder and task-specific adapters, achieving strong performance on both classification and regression tasks.

This paper introduces a two-stage framework combining 3D UNet and 3D GCN for accurate Couinaud segmentation in tumor-affected livers, achieving a mean Dice score of 0.828 on blind testing.

This paper introduces a computational framework to quantify transcriptomic damage across tissues and conditions, revealing age-associated increases and demonstrating that interventions can alter tDamAge.

This paper evaluates agentic AI systems on three scientific tasks, finding that human-led AI-assisted work outperforms autonomous systems, with one AI matching human performance in property prediction.

This study develops a machine learning framework to predict CRC stage transitions using tissue-derived transcriptomic signatures and evaluates their performance in blood-based datasets.

This paper benchmarks prior bias in deep learning models for de novo peptide sequencing using a novel framework called DeNovo-PBI.

This paper introduces a method to control false discovery rates in deep-learning-based de novo peptide sequencing, improving reliability of peptide identifications.

The paper introduces two metrics to distinguish biological and interventional changes in longitudinal data using ADNI cognitive measurements.

The paper presents a framework for designing active enzymes from functional descriptions, achieving high catalytic activity in synthesized designs.

SentryPath is a mechanistic simulator that cross-validates across 13 oncology trials and forecasts trial outcomes with 3.7% mean error in NSCLC.

The paper evaluates cross-validation strategies for RNA secondary structure prediction, showing that current methods bias results and proposing homology-aware strategies for fair comparison.

OpenGerminal replaces PyRosetta with open-source tools and benchmarks IgLM versus AbLang1 in the Germinal antibody design pipeline, achieving higher cofolding pass rates with equivalent structural confidence.

The paper introduces a unified model for biological prediction that integrates geometry, mechanism, and causal structure, achieving lower error than Euclidean and local graph baselines in synthetic studies.