Analytical strategies for discovery and replication of genetic effects in pharmacogenomic studies

Abstract

In the past decade, the pharmaceutical industry and biomedical research sector have devoted considerable resources to pharmacogenomics (PGx) with the hope that understanding genetic variation in patients would deliver on the promise of personalized medicine. With the advent of new technologies and the improved collection of DNA samples, the roadblock to advancements in PGx discovery is no longer the lack of high-density genetic information captured on patient populations, but rather the development, adaptation, and tailoring of analytical strategies to effectively harness this wealth of information. The current analytical paradigm in PGx considers the single-nucleotide polymorphism (SNP) as the genomic feature of interest and performs single SNP association tests to discover PGx effects – ie, genetic effects impacting drug response. While it can be straightforward to process single SNP results and to consider how this information may be extended for use in downstream patient stratification, the rate of replication for single SNP associations has been low and the desired success of producing clinically and commercially viable biomarkers has not been realized. This may be due to the fact that single SNP association testing is suboptimal given the complexities of PGx discovery in the clinical trial setting, including: 1) relatively small sample sizes; 2) diverse clinical cohorts within and across trials due to genetic ancestry (potentially impacting the ability to replicate findings); and 3) the potential polygenic nature of a drug response. Subsequently, a shift in the current paradigm is proposed: to consider the gene as the genomic feature of interest in PGx discovery. The proof-of-concept study presented in this manuscript demonstrates that genomic region-based association testing has the potential to improve the power of detecting single SNP or complex PGx effects in the discovery stage (by leveraging the underlying genetic architecture and reducing the multiplicity burden), and it can also improve power in the replication stage.

A comparison of multiple testing adjustment methods with block-correlation positively-dependent tests

“…When considering the FDR, we use m = 2000 features—arbitrarily a high number, which involves substantial-but-manageable computational expense in dealing with , and which is close to the number of features in a microRNA study [ 19 , 20 ]. When considering the FWER, we use m = 100 features, arbitrarily a low number, but one which is reasonable in a pharmacogenomics PGx subgroup analysis [ 21 ] or methylation quantitative trait loci study [ 22 ]. We fix the percent differential abundance and number of features thus in the simulation, not because in practice we could assume the same percentage differentially abundant or the same number of features in all studies, but rather because our focus is on how the degree of dependence and magnitude of differential abundance (and not percentage differentially abundant or number of features) affect the performance of the multiplicity adjustment methods considered here.…”

Textbook of Personalized Medicine

“…Most of the ongoing SNP related studies are biased deliberately towards coding regions and the data generated from them are therefore unlikely to refl ect genome wide distribution of SNPs (Katara 2014 ). Single SNP association testing is suboptimal given the complexities of the clinical trial setting, including: (1) relatively small sample sizes; (2) diverse clinical cohorts within and across trials due to genetic ancestry (potentially impacting the ability to replicate fi ndings); and (3) the potential polygenic nature of a drug response (Kohler et al 2014 ). The authors of this proof-of-concept study propose a shift in the current paradigm to consider the gene as the genomic feature of interest in pharmacogenomics discovery.…”