A Regression Framework for Assessing the Covariate Effects on the Reproducibility of High-Throughput Biological Experiments

• Activity Number: 499
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Qunhua Li*+
• Companies: Penn State
• Keywords: genomics ; reproducibility ; copula ; ordinal data ; high-throughput data ; tail dependence

Abstract

Fidelity of discoveries from high-throughput biological experiments is often affected by many experimental or data analytical factors. In this work, we propose a regression framework to assess the covariate effect of operational factors on the reproducibility of discoveries from high-throughput experiments. Unlike the usual measures of agreement, our reproducibility measure focuses on the rank consistency across replicates for the top-ranked candidates -- the primary interest of high-throughput experiments. Based on this measure, we propose a regression framework to assess the covariate effect on the reproducibility. Our method allows one to characterize the simultaneous and independent effects of covariates on reproducibility and to compare reproducibility while controlling for potential confounding variables. It is related to cumulative link models and certain copula models.

Using simulations, we show that our method produces correctly calibrated type I error and is more powerful in detecting the difference in the reproducibility of top-ranked candidates than usual measures of agreement. We illustrated the effectiveness of our methods using ChIP-seq and microarray studies.