variant effect prediction

The paper introduces Tν, a metric to quantify directional asymmetry in amino acid substitutions using kernel regression on MSAs, and applies it to human glutamate transporters to identify high-tension sites and pathogenic variants.

This study benchmarks DNA foundation models like Evo2, revealing blind spots in variant-effect prediction and highlighting reliance on sequence regularities over biological constraints.

BICEP extends its Bayesian model to include indels and copy number variants for rare variant prioritization in pedigrees, showing comparable performance to the original single nucleotide variant model.

HAETAE predicts tissue-specific histone modifications with high accuracy using a 5-base framework and long-read sequencing data.

The paper investigates whether genome size, replichore organization, and nucleotide skew predict bacterial growth rates, finding that skew improves predictive accuracy.

The paper identifies functional variants and VNTRs at the 5p15.33 locus using fine-mapping, MPRA, and CRISPRi screens, revealing their role in cancer susceptibility through antagonistic pleiotropy.

The paper develops a framework to train supervised models on MAVE data by accounting for experimental protocol differences, improving variant effect prediction across datasets.

AMR-GNN predicts antimicrobial resistance using a graph neural network framework that integrates multiple genomic representations from whole-genome sequencing data.

The paper develops machine learning models to predict pathogenicity of BRCA1 and BRCA2 variants using cancer genomic profiles, improving classification of variants of uncertain significance.

This paper reviews sparse autoencoders applied to biological foundation models, finding they recover interpretable biological features but lack experimental validation of mechanistic understanding.

The paper introduces Gene Portals, a framework that integrates clinical, functional, and structural data for rare disease variant classification, demonstrating its utility across five gene-specific portals.

The paper identifies rare noncoding variants associated with Parkinson's disease using multi-omic data from over 3.3 million nuclei.

The paper identifies unreliable confidence metrics in AlphaFold2 predictions for fold-switching proteins and introduces a screening method to detect false positives.

EvoStructCLIP predicts missense variant effects by integrating 3D structure and evolutionary data, showing transferability across gene-specific functions and biological tasks.

This study identifies 6mA demethylases as key regulators of virulence in plant-parasitic nematodes and shows their potential as targets for sustainable control.

The paper uses deep-learning models to identify functional motifs in intrinsically disordered proteins by analyzing pathogenic variants from ClinVar.

BOTANIC-0 models are plant genomic foundation models pretrained on diverse plant genomes, achieving competitive performance in genomic tasks like variant effect prediction.

The paper introduces the Intolerome Database, a curated resource of genes essential for human viability, to help prioritize genomic variants in cases of stillbirth, pregnancy loss, and neonatal death.

This study identifies DMT3 as a 6mA demethylase in unicellular eukaryotes and shows its role in maintaining transcriptional and developmental integrity by preventing ectopic 6mA deposition at TSS.

This study performs a GWAS on a diverse grapevine collection to identify genetic determinants of berry size and cluster architecture traits, finding significant SNPs and candidate genes.

The paper quantifies how single nucleotide variants affect NKX2.1 transcription factor binding and uses neural networks to predict disease-causing variants in regulatory regions.

The paper develops a machine-learning model to predict how structural variants disrupt 3D chromatin structure in congenital heart disease and validates its predictions using induced pluripotent stem cells.

The paper performs a genome-wide CRISPR activation screen to identify genes modulating PrPC abundance, finding 531 genes with significant effects and validating 45 of them.

The paper evaluates nine LLPS predictors across diverse protein landscapes, finding significant variability in performance and highlighting the need for diverse benchmarking data.

The paper evaluates deep learning architectures for high-resolution chromatin accessibility prediction, finding that early feature extraction, not downstream architecture, primarily determines performance.

This study investigates mutation dynamics in Komagataeibacter intermedius under agitation stress, finding that non-structural variations, not structural ones, drive the loss of bacterial cellulose production.

D2D improves protein mutation effect predictions using evolutionary constraints and a language model, outperforming existing methods in stability, binding, and epistasis.

SPC improves population structure adjustment in genomic analysis by capturing fine-scale patterns better than principal components, reducing p-value inflation and improving rare variant associations.

STRUMP-I improves pMHC-I binding prediction by using force-field energy features, outperforming sequence-based methods on underrepresented MHC alleles and showing synergy with them.