In this talk, I will discuss how the analysis of a clinical genomics study has lead us to a new theoretical result in statistics. A major clinical study has been underway for the past six years that is attempting to understand the molecular basis of adverse inflammatory reactions to severe trauma. Genome-wide gene expression has been measured longitudinally in a cohort of individuals who have undergone severe blunt force trauma. One of the main goals of this study is to characterize gene expression trajectories to understand and possibly predict the patients' outcomes. Through real data issues encountered in this study, we have been lead to a theoretical solution to an important open problem in statistics, namely how to deal with dependence in large-scale data sets.

Capturing Heterogeneity in Gene Expression Studies by Surrogate Variable Analysis