COMPUTATIONAL ANALYSIS OF MISSENSE VARIANTS LINKED TO INTER-INDIVIDUAL VARIABILITY IN HUMAN DISEASES AND DRUG TREATMENT

Abstract

Interpretation of genomic variation is a key challenge in the biomedical field that is complicated by the recent avalanche of genomic data containing many variants of unknown significance (VUS). Here, we leveraged data from sequence variation repositories and used computational tools to analyse missense VUS that may modulate health or treatment outcomes. By in silico analyses, we discovered rare germline variants that could ablate or generate peroxisome targeting signals in disease-relevant proteins that were later verified experimentally. Similarly, we also identified novel population-specific common variants in cytochrome P450 pharmacogenes (CYP) for further biochemical characterization in understudied populations. We then built a machine learning-based classifier that categorizes CYP missense VUS into disease, neutral or pharmacogenomics-associated (PGx) groups that outperformed a dummy classifier in ternary classification and oft-cited tools (eg: SIFT, PolyPhen-2) for binary classification. Overall, our work has contributed to the further understanding and discrimination of missense VUS.