Statistical Modeling for Multiplex RNAi Screen Data Analysis

Abstract

Multiplex RNAi screen is an emerging tool for functional genomics. Most analysis methods presently available for Multiplex RNAi screen are based on single hairpin data. These approaches have serious limitations. They do not account for the redundancies in genome-scale libraries. Thus it is difficult to detect genes with modest but consistent effect. In addition, contradictory conclusions might be reached based on enriched and depleted hairpins for the same gene. Therefore, we propose the RNAi Set Enrichment Analysis (RSEA) framework based on the gene set enrichment analysis framework that will take multiple hairpins into consideration in accessing the gene effect on drug response. The gene set enrichment analysis has been widely used in gene expression microarray study to test whether a certain biological pathway is activated under some treatment. However this method is rarely used in RNAi screen studies. With the RSEA method, we evaluate and compare the performance of different RNAi level statistics, RNAi set statistics and significance assessment choices. Besides these, to model the silencing efficiency and off target effect of RNAi knockdown, we propose Structural Equation Modeling (SEM) with latent variables for RNAi screen data analysis. SEM is intuitive for biological researchers with its path diagrams. In addition, the latent SEM contains the repeated measures ANOVA, both the univariate and the multivariate approaches, as special cases. Our simulation studies revealed that the latent SEM has comparable statistical power to RSEA method when the hairpin off target effect is modest. While the adoption of the SEM to existing experimental data is hampered by the modest sample size, we are able to verify the RSEA method by applying them towards real data generated from our experiments. The result shows that RSEA can successfully identify positive genes whose effects have been validated by the follow-up confirmatory experiments.