A Survival Copula Mixture Model for Comparing Two Genomic Rank Lists

Abstract

Analyses of high-throughput genomic data often lead to ranked lists of genomic loci. How to characterize concordant signals between two rank lists is a common problem with many applications. One example is measuring the reproducibility between two replicate experiments. Another is to characterize the interaction and co-binding between two transcription factors (TF) based on the overlap between their binding sites. As an exploratory tool, the simple Venn diagram approach can be used to show the common loci between two lists. However, this approach does not account for changes in overlap with decreasing ranks, which may contain useful information for studying similarities or dissimilarities of the two lists. The recently proposed irreproducible discovery rate (IDR) approach compares two rank lists using a copula mixture model. This model considers the rank correlation between two lists. However, it only analyzes the genomic loci that appear in both lists, thereby only measuring signal concordance in the overlapping set of the two lists. When two lists have little overlap but loci in their overlapping set have high concordance in terms of rank, the original IDR approach may misleadingly claim that the two rank lists are highly reproducible when they are indeed not. In this article, we propose to address the various issues above by translating the problem into a bivariate survival problem. A survival copula mixture model is developed to characterize concordant signals in two rank lists. The effectiveness of this approach is demonstrated using both simulations and real data.