Abstract 16981: A Machine Learning Model Increases Genetic Confirmation of UKB Participants With Suspected Monogenic Cardiovascular Disease

Abstract

Introduction: Monogenic cardiovascular diseases (MCVDs) are collectively common yet highly underdiagnosed. Genetic testing for pathogenic variants can facilitate diagnosis and familial screening of MCVDs. Yet, most genetic variants within MCVD-associated genes currently annotated in ClinVar are variants of uncertain significance (VUS). VUS clinical utility is limited as VUS cannot be used for genetic confirmation of disease or for familial cascade screening. We previously developed a disease- and tissue-specific machine learning model for reclassification of VUS (Cardiovascular Disease-Pathogenicity Predictor [CVD-PP]). We sought to determine if use of CVD-PP on VUS identified through whole exome sequencing (WES) would increase genetic diagnosis of patients suspected of having a MCVD in the UK Biobank (UKB) cohort. Methods: We filtered for variants in 47 genes associated with 5 MCVDs (cardiomyopathies, channelopathies, aortopathies, cardiac amyloidosis, and familial hypercholesterolemia) in UKB participants with WES data. Participants harboring an American College of Medical Genetics confirmed ClinVar pathogenic or likely pathogenic (P/LP) variant in one of these genes were excluded. Phenotyping for presence of MCVDs was performed using EHR, clinical, cardiac imaging, and ECG data. CVD-PP was utilized to predict the pathogenicity of variants that were classified as VUS, had conflicting interpretations, or were absent from ClinVar. Results: Of participants with QC’ed WES data in UKB (n=452,933), 34.7% (n=157,030) had expression of a MCVD. Of these, 0.51% (n=799) carried a P/LP variant in a MCVD gene. Of the remaining P/LP variant-negative participants expressing disease (n=156,231), 4.1% (n=6,346) carried a VUS or non-ClinVar variant that was predicted pathogenic by CVD-PP. Conclusion: We demonstrate the potential utility of using a disease-and tissue-specific computational tool, CVD-PP, for deeper characterization of VUS in MCVD genes. Prioritization of CVD-PP-predicted pathogenic variants increases yield of genetic diagnosis in patients with evidence of clinical MCVDs from 0.51% using only ClinVar P/LP variants to 4.6% in UKB. This work provides a potential framework for increasing overall diagnostic yield of MCVDs.