Sample Sizes Associated with a Choice of Normalization and Test Statistical Methods for Differential Gene Expression Analysis in RNA-Seq Studies

• Activity Number: 403 - SPAAC Poster Competition
• Type: Topic Contributed
• Date/Time: Tuesday, July 30, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Xiaohong Li* and Nigel G.F. Cooper and Timothy E O'Toole and Eric C. Rouchka
• Companies: University of Louisville and University of Louisville and University of Louisville and University of Louisville
• Keywords: Normalization; RNA-seq; Differential gene expression; Sample size

Abstract

Normalization of high-throughput RNA sequencing (RNA-seq) data and statistical tests are essential steps for identifying differentially expressed genes (DEGs). The most commonly used normalization methods are TMM (Trimmed-Mean M-values), RLE (Relative Log Estimate) and UQ (Upper quartile) normalization. The common statistical tests are a Wald test from DESeq2 and an exact test from edgeR. Although several comparative studies reported that DESeq is more conservative than edgeR, both failed to maintain a false discovery rate below a nominal level of 0.05. Recently, we observed that a UQ-pgQ2 normalization combined with an exact test from edgeR has a better specificity for DEG analysis using benchmark MAQC data and simulated data for small sample sizes/replicates. However, for a larger sample size, it remains uncertain if an exact test performs better than a Wald test? To address this question, we evaluated the performance of these methods combined with two tests. We observed that a Wald test perform better than an exact test in controlling for false positives when sample sizes are large. However, an exact test combined with UQ-pgQ2 is best choice when sample sizes are small.