Nadav Brandes

The paper evaluates AI models for predicting pathogenic variants, finding that performance is inflated by genomic heterogeneity in pathogenic variant prevalence.

P-KNN improves variant classification by jointly calibrating multiple pathogenicity tools, achieving better calibration and evidence strength than individual tools.

The paper uses deep mutational scanning and structural analysis to study the functional effects of missense variants in the P2RY8 receptor and improves variant effect predictions.

The paper investigates why carriers of pathogenic variants in monogenic metabolic conditions show variable disease severity, finding support for variant effect sizes, polygenic backgrounds, and epistasis.

PWAS Hub identifies gene-disease associations for 99 common diseases using proteome-wide association analysis, with sex-specific insights and comparison to GWAS results.

The paper uses single cell RNA sequencing to assess the effects of protein-altering variants in three genes, showing that single cell expression signatures are more effective than bulk assays and computational models for predicting pathogenicity.

This paper evaluates covariate selection in GWAS for 36 UK-Biobank traits, proposing guidelines to improve power and reduce computational burden.

This paper uses a deep protein language model to predict the functional impact of all human missense variants, outperforming existing methods and providing isoform-specific annotations.

A ~16M-parameter protein language model, pretrained on ~106M UniRef90 proteins by denoising sequences and Gene Ontology annotations, rivals larger TAPE models using linear-cost global attention.

This study uses GWAS and PWAS on UK Biobank data to identify 137 cancer risk loci, highlighting recessive effects and cancer-type specificity.

The paper constructs human-specific amino-acid substitution matrices from genetic variation data to better represent proteomic variation within the human population and reveal new evolutionary insights.

PWAS detects protein-coding genes associated with phenotypes by analyzing proteomic context and using machine learning, showing improved detection of causal genes and new associations.

The paper introduces a machine learning framework to identify cancer driver genes by analyzing the functional impact of somatic mutations, finding 593 genes with significant harmful mutation bias.

The paper analyzes viral genome size, gene overlapping, and capsid properties across 100 viral families, finding that gene overlapping is tightly bounded and likely contributes to gene novelty rather than genome compression.

ASAP is a machine-learning framework for predicting residue-level protein properties, demonstrating high performance in predicting protein cleavage sites.