Discover, rate, and discuss biology preprints.

This paper introduces a graph transformer model for drug-target binding affinity prediction, showing improved performance on benchmark datasets.

This paper uses a phylogenetic approach to analyze bradyzoite differentiation in Toxoplasma gondii and identifies conserved and novel proteins involved in this process.

SPaCeD improves heavy-light chain pairing accuracy in spatial BCR-seq by integrating expression matrices with spatial distances via optimal transport.

This paper introduces a mechanotransduction-aware causal omics framework to identify disease genes in mechanically active tissues by modeling mechanotransduction as part of the causal structure rather than a secondary covariate.

The paper improves target identification in chemical proteomics by using Fisher's method to aggregate peptide-level data instead of protein-level data.

CD8scape predicts viral escape from CD8+ T cell response using HLA genotype and within-host sequence variation to identify immune escape potential.

PEACE improves effector protein prediction by using contrastive embeddings and prototype-aware training, outperforming EffectorP 3.0 on fungal and fungi+oomycete datasets.

The paper develops a kernel matrix completion method integrating topological and spectral features to classify yeast proteins, achieving improved ROC performance on incomplete data.

This paper quantifies the limitations of drug-blind prediction in cancer DRP models, showing that generalization relies on mechanistic overlap and that performance is more about dataset behavior than model behavior.

The paper expands P-NET, a deep learning framework that uses biological pathways for genomic and transcriptomic predictions, showing competitive performance and interpretability.

The paper searches a petabase-scale dataset to discover new Papillomaviruses, identifying 383 novel types across 105 host species and highlighting biodiversity hotspots.

This study benchmarks kinase bioactivity models across splitting strategies and protein representations, finding that Random Forest with Morgan fingerprints outperforms neural models in scaffold and target-based evaluations.

The paper develops TargetTrace, a resource linking drug targets to pharmacodynamic biomarkers, compiling 71,323 target-biomarker relationships for oncology drug development.

The paper compares machine learning models for predicting single-cell protein expression from RNA-seq data to mRNA levels, finding that models outperform mRNA as a proxy.

HMCVelo computes hydroxymethylation velocity in single cells using a deterministic ODE model, outperforming RNA velocity in trajectory inference for murine cortical cells.

GeomMotif evaluates protein generation models' ability to preserve arbitrary structural fragments, revealing differences in performance based on structural and physicochemical context.

This paper introduces Surleton, a framework for inverse protein folding that jointly models backbone and surface geometry, improving sequence recovery and predictive confidence on benchmark datasets.

This study benchmarks single-cell foundation models for data integration using real-world RNA-seq data and finds that fine-tuned scGPT_CP performs best in technical correction while scVI is the top overall performer.

CoPro identifies spatially coordinated gene expression programs across tissues, revealing distinct cellular states and their interactions in colon, brain, liver, and kidney.

This paper introduces fluProfiler, a framework using viral sequences to predict antigenic relationships and improve vaccine strain selection by reducing reliance on time-consuming HI assays.