Automating the Interpretation of Pharmacogenetic Data

Abstract

Genetic polymorphism contributes significantly to the wide inter-individual variability in drug response, affecting both efficacy and toxicity. It has been estimated that more than 90% of the United States population has at least one clinically actionable pharmacogenetic (PGx) variant that affects their response to medication. Variation in the enzymatic activity of pharmacogenes is defined by star alleles (haplotypes) comprised of single nucleotide variants, small insertion-deletions, and large structural variants (SVs). In this dissertation, I detail the development and application of Stargazer, a novel SV-aware algorithm that can call star alleles in various polymorphic pharmacogenes from next-generation sequencing (NGS) data. When developing Stargazer, I selected the clinically important CYP2D6 gene as a starting point because the enzyme it encodes metabolizes approximately 25% of drugs, and it is one of the most difficult genes to genotype in the human genome. To assess the performance of Stargazer, I utilized targeted sequencing data of 32 ethnically diverse trios that were genotyped for CYP2D6 by multiple orthogonal methods. Next, I applied Stargazer to targeted sequencing data from human liver tissues (N>300) with deep phenotyping data to evaluate Stargazer’s predictive power for CYP2D6 mRNA expression, protein abundance, and enzyme activity. Finally, I extended Stargazer to 28 key pharmacogenes using whole genome sequencing data from 70 reference samples that were robustly characterized by several PGx testing assays. Taken together, this dissertation demonstrates that the combination between NGS and Stargazer offers a feasible path for accurate PGx analysis and prediction of individual drug responses. This approach will be increasingly useful in clinical practice, particularly as whole genome sequencing and targeted panel sequencing become more widely available.