AlphaCluster: Coevolutionary driven residue-residue interaction models enable quantifiable clustering analysis of de novo variants to enhance predictions of pathogenicity

Obiajulu J, Kuang R, Zhong G, Hagen J, Shu C, Chung WK, Shen Y

Research Square, 2022.

Lab members marked as bold

Abstract

Missense variants have highly variable effects and effect size, which often makes it challenging to distinguish pathogenic and non-pathogenic variants and subsequently implicate new genes for disease association in studies of de novo and inherited rare variants. Importantly, missense variants can be the sole molecular mechanism for some genetic disorders, and so statistical approaches tailored for the analysis of missense variants are critical. Analysis of the clustering of missense variants is a promising approach which leverages the fact that missense variants in protein domains often have similar effects on function. Here we describe a new clustering analysis approach, AlphaCluster, a statistical method which quantifiably analyzes the spatial clustering of de novo variants by mapping missense residues onto the protein tertiary structure. We show that our approach can quantify the evidence supporting pathogenic missense variants and increase the power to detect clustering when compared to available genomic clustering tools. Using AlphaCluster, we identified genes newly implicated in autism spectrum disorder and neurodevelopmental disorders (NDD). We also apply AlphaCluster to protein complexes and detect an association between the gamma aminobutyric acid receptor complex (GABA-A alpha-1/beta-2/gamme-2 receptor).