memVar: an integrated resource for interpreting human membrane protein variation in molecular and disease contexts

Abstract

Interpreting membrane protein variation requires connecting clinical evidence to the membrane environments and molecular interactions that govern protein functions. Here, we introduce memVar (https://memvar.sunnylab.org), an integrated database of 10,906,688unique single-nucleotide variants from ClinVar, COSMIC, and gnomAD across 7,732 human membrane-associated proteins. memVar reconciles genomic, transcript, and protein coordinates to link variants with membrane topology, bilayer depth, lipid contacts, functional sites, partner-specific interaction interfaces, and variant-effect predictions, alongside disease, tissue-expression and regulatory annotations. Coordinated sequence, structure and variant views retain source provenance and distinguish clinical assertions, experimental observations, and computational predictions. Our analysis reveals that pathogenic/likely pathogenic missense substitutions are enriched in transmembrane segments and predicted protein-protein interaction sites relative to benign/likely benign substitutions, with pooled odds ratios of 2.47 and 6.67, respectively. IntAct-derived interaction loss or decrease was experimentally recorded for 57.9% of pathogenic/likely pathogenic substitutions versus 7.4% of benign/likely benign substitutions. CFTR examples contextualize the established G551D gating defect and prioritize W202R, a variant of uncertain significance, for tests of protein maturation, surface abundance and channel activity. By connecting variant records to specific molecular questions, memVar potentially supports mechanism-oriented investigation and prioritization of functional experiments.