Variant annotation across homologous proteins (“Paralogue Annotation”) identifies disease-causing missense variants with high precision, and is widely applicable across protein families

Abstract Background Distinguishing pathogenic variants from those that are rare but benign remains a key challenge in clinical genetics, especially for variants not previously observed and characterised in humans. In vitro and in vivo functional characterisation are typically resource intensive, and model systems may not accurately predict influence on human disease. Many in silico tools have been developed to predict which variants are disease-causing, but typically lack precision. Here we demonstrate the applicability of a framework, called Paralogue Annotation, that draws on information from previously-characterised variants in homologous proteins to predict whether variants in a gene of interest are likely disease causing. Methods We assessed the performance of Paralogue Annotation through three orthogonal approaches: (1) comparison to established in silico variant prediction tools using 47,360 missense variants from ClinVar across 3,524 genes representing a broad range of diverse protein classes, by calculating precision and sensitivity; (2) evaluation against large-scale functional assays of variant effect in TP53 and PPARG ; and (3) comparing odd ratios calculated from case-control association tests for inherited cardiac arrhythmia syndromes, and neurodevelopmental disorders with epilepsy, stratifying variants by Paralogue Annotation. Results Paralogue Annotation correctly annotates 4,328 ClinVar pathogenic variants, with 245 false positives, yielding a precision of 0.95. This increases to 0.99 with more stringent annotation parameters (requiring greater conservation of amino acids between annotated orthologues) at the expense of sensitivity. Compared to established tools, Paralogue Annotation has higher precision for identification of pathogenic variants, albeit with lower sensitivity across diverse test sets. Extending the technique by transferring annotations between homologous protein domains, rather than full-length protein paralogues, increases sensitivity. Rare variants predicted pathogenic by Paralogue Annotation were more strongly disease-associated (increased odds ratio) than unstratified rare variants for six out of eight genes tested with case-control cohort approaches. Conclusions Paralogue Annotation has high precision for detection of pathogenic missense variants, outperforming in silico methods where data are available to make a prediction. As the number of characterised variants increases in reference datasets such as ClinVar, Paralogue Annotation will further increase in sensitivity and applicability.