DeMAG predicts the effects of variants in clinically actionable genes by integrating structural and evolutionary epistatic features

Abstract

Despite an increasing use of genomic sequencing in clinical practice, interpretation of rare genetic variants remains challenging even in well-studied disease genes, resulting in many patients with Variants of Uncertain Significance (VUSs). Computational Variant Effect Predictors (VEPs) are currently used to provide valuable evidence in variant classifications, but they often misclassify benign variants, contributing to potential misdiagnoses. Here, we developed Deciphering Mutations in Actionable Genes (DeMAG), a supervised classifier for interpreting missense variants in actionable disease genes with improved performance over existing VEPs (20% decrease of false positive rate). Our tool has balanced specificity (82%) and sensitivity (94%) on clinical data, and the lowest misclassification rate on putatively benign variants among evaluated tools. DeMAG takes advantage of a novel epistatic feature, the partners score, which is based on evolutionary and structural partnerships of residues as estimated by evolutionary information and AlphaFold2 structural models. The partners score as a general framework of epistatic interactions, can integrate not only clinical but functional information. We anticipate that our tool (demag.org) will facilitate the interpretation of variants and improve clinical decision-making.

A Functional Map of the Human Intrinsically Disordered Proteome

“…The sequence-structure-function paradigm in molecular biology posits that the amino-acid sequence of a protein encodes its three-dimensional structure, which determines the function of the protein. The close relationships between sequence, structure, and function are routinely exploited to infer function from sequence or structural data (Ashburner et al 2000; Lee et al 2007; Radivojac et al 2013; Sanderson et al 2023; Yu et al 2023b), trace the evolutionary history of protein-protein interactions (Steube et al 2023), design de novo proteins with desired folds or functions (Huang et al 2016; Kuhlman & Bradley 2019; Yeh et al 2023), and predict the pathogenicity of sequence variants in the human genome (Adzhubei et al 2010; Frazer et al 2021; Hopf et al 2017; Luppino et al 2023). Indeed, structural information recovered from amino-acid sequence alignments is central to state-of-the-art protein structure prediction methods (Baek et al 2021; Jumper et al 2021).…”