Detecting pharmacokinetic and pharmacodynamic covariates from high-dimensional data

With the rapid evolution of technologies capable of generating high-dimensional data sets such as those from the ‘omic’ platforms commonly encountered during pharmacogenetic/genomic clinical trials, there is need for computationally efficient methodologies capable of integrating that information into the drug development pipeline; however, the computational cost of identifying covariates and interactions through traditional parametric statistical approaches has impeded their utilization for these large data sets. Within the context of population pharmacokinetic/pharmacodynamic modeling, the potential for detecting interactions on such data sets is of great interest: Specifically, the applications of interactions in this context would be the creation of more comprehensive and biologically sound covariate models, leading to better prediction of individual values for PK/PD parameters of interest, and moving one step closer to the goal of personalized medicine. However, there are currently no commercially available software packages, or computational approaches, that can handle covariate interaction detection or model synthesis at a genome scale. Thus, the most immediate and tractable benefit from such interaction analyses at this scale would be the identification of the most informative subset of predictors that could be used for ‘formal’ covariate model synthesis. This chapter will provide a discussion on the following topics of interest in this area: A general discussion on covariates and interactions; specific challenges and opportunities that arise when large datasets are considered; search metrics that are applicable on high-dimensional data sets; and a justification for the need to distinguish between covariate detection and formal covariate model synthesis in this context.