Predrag Radivojac*
Khoury College of Computer Sciences, Northeastern University, Boston MA, USA
predrag [at] northeastern.edu
Abstract
Resolving variants of uncertain significance (VUS) is a key challenge in genomic medicine, with about most of the variants in ClinVar classified as VUS. This limits the clinical utility of genetic testing for patients with suspected Mendelian disease and is a hindrance to improving human health and well-being. Two major sources of evidence hold promise for resolving VUS at scale: (1) variant effect predictors, which generate computational scores for any variant, and (2) high-throughput experimental assays, which directly measure the impact of variants on gene function. Realizing the clinical potential of these evidence types requires systematic data generation, rigorous calibration, and data sharing. We will mostly focus on new calibration strategies for computational and experimental evidence, and demonstrate their combined power to dramatically reduce the burden of VUS.
Towards the goal of reducing VUS, we generated experimental measurements for over 60,000 variants across ten genes using multiplexed assays and curated about 200,000 existing variant effect measurements across 30 genes from the literature. Using the new calibration framework, we developed a scalable classification workflow relying solely on experimental and predictive evidence that enables reclassification of about 75% VUS across these genes as pathogenic or benign, with error below 1%. The framework was also applied prospectively. Among more than 90,000 variants not yet observed in clinical databases, 62% carried sufficient evidence to be "preclassified" before ever appearing in a patient record. This approach establishes a blueprint for how coordinated data generation, calibration, and open dissemination can transform genomic medicine.
Keywords: calibration strategies, uncertain significance variants