Finding Surrogate Controls in Pathway Enrichment Analysis using RNA-Seq Expression

摘要

In this work, we present a method of imputing controls for test samples that do not have matching normal samples. We call the imputed one a subject-specific surrogate control. We study if this method of using a subject-specific surrogate control produces any better outcome than the conventional method of using the population average as the sole common control for all test samples. For evaluation, we apply our method to publicly available gastric and breast cancer data from the Cancer Genome Atlas (TCGA). First, we separated the test samples into two subgroups, one having elevated level of Epithelial-Mesenchymal Transition (EMT) and the rest. Second, we perform the Kaplan-Meier (KM) survival analysis for the two identified subgroups. In both cancer cases, we discover that EMT elevated population has a poorer prognosis than the subgroup having non-elevated EMT level. EMT elevated cancer patients having a poorer prognosis has been well established in gastric cancer, but the same observation for the breast cancer case has not. We hypothesize that some subtle differences that cannot be revealed using the population control can be discovered if subject-specific surrogate method is used, thus suggesting a potential merit for our newly suggested subject-specific control imputation method.