Project Summary Insufficient knowledge and throughput to interpret pathogenicity of genetic variants identified by next generation sequencing (NGS) is a major bottleneck for genomic medicine implementation. The American College of Medical Genetics and Genomics and Association for Molecular Pathology (ACMG/AMP) guidelines identify high-confidence pathogenic and likely pathogenic variants but are limited in scalability. Many variants are classified as variants of uncertain significance by the ACMG/AMP guidelines without an indication of which of these variants are more or less likely to be pathogenic, leading to inappropriate medical treatment. Hence, I propose to develop standardized quantitative approaches to improve our ability to interpret genomic variations accurately at high-throughput. In-silico tools are commonly used to assign variant pathogenicity based on conservation, but their predictive accuracy is limited. The current methods have not been calibrated across genes, and the same pathogenicity score does not infer the same likelihood of pathogenicity across different genes. In this proposal, 1) I aim to recalibrate the pathogenicity scores incorporating gene-specific features making the pathogenicity scores more comparable across genes, and improve the accuracy of pathogenicity predictions using advanced deep neural network models and functional data from saturation mutagenesis studies. 2) I aim to quantify the ACMG/AMP variant classification and provide probability of variant pathogenicity for clinically relevant genes using advanced supervised learning and leveraging a large case- control cohort. The improved computational predictions (Aim 1) will refine variant prioritization for downstream analyses and strengthen the computational evidence used in the ACMG/AMP guidelines. The estimated probability of variant pathogenicity based on ACMG/AMP guideline (Aim 2) will improve communication between laboratories, health care providers and patients about genetic test results.