Per-Gene Normalization Method (UQ-PgQ2) Improves the Specificity for the Analysis of Differential Gene Expression in RNA-Seq Data

• Activity Number: 254 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Xiaohong Li* and Nigel G.F. Cooper and Dongfeng Wu and Eric C. Rouchka and Shesh N. Rai
• Companies: University of Louisville and University of Louisville and University of Louisville and University of Louisville and University of Louisville
• Keywords: Normalization; RNA-seq; DE gene; UQ-pgQ2; DESeq2; edgeR

Abstract

Sample normalization is an essential step with considerable impact on the analysis of differentially expressed genes (DEGs) in high-throughput RNA sequencing (RNA-seq) experiments. Although there are numerous methods for normalizing read counts to allow for comparative analysis, it remains a challenge to maintain the actual false discovery rate (FDR) below a nominal level. To specifically address this issue, we developed an UQ-pgQ2 normalization method, which is the median per-gene normalization (pgQ2) following the upper-quantile per-sample global scaling. In this work, we compared the UQ-pgQ2 method with three most commonly used methods (UQ, DESeq2 and edgeR)using two benchmarked Microarray Quality Control (MAQC) RNA-seq datasets. An additional within-group comparison based on the publically available TCGA datasets was used to further assess normalization methodologies. The results show that the UQ-pgQ2 method combined with a Wald test from DESeq2 has the smallest number of false positives given a FDR cutoff of 0.05. We conclude that our method outperforms UQ, DESeq2 and edgeR by improving the DEG specificity.