A Comparison of Combined P-value Methods for Gene Differential Expression Using RNA-Seq Data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-Seq data analysis is crucial for understanding biological processes.
• Gene-level analysis of RNA-Seq data can be inaccurate due to read distribution heterogeneity.
• Regional analysis, such as exon-level, combined with p-value aggregation may improve DE gene detection.

Purpose of the Study:

• To evaluate the performance of various p-value combining methods for detecting differentially expressed genes (DEGs) in RNA-Seq data.
• To compare Fisher's, Z-transform, Weighted Z-test, Minimum P-value, Logit, and Weighted-sum methods.
• To assess the effectiveness of these methods on liver, kidney, and MAQC datasets.

Main Methods:

• Applied six widely-used p-value combining methods to publicly available RNA-Seq datasets.
• Analyzed performance based on the detection of true DE genes and true non-DE genes.
• Investigated the impact of weights in the Weighted Z-test and potential issues with p-value independence.

Main Results:

• The Weighted Z-test demonstrated superior performance in detecting true DE genes on liver and kidney datasets.
• On MAQC datasets, methods performed similarly, with a slight advantage for Weighted Z-test and Fisher's method in detecting DE genes.
• The Weighted-sum method excelled at identifying true non-DE genes, indicating an inverse performance relationship.

Conclusions:

• The Weighted Z-test is effective for identifying true DE genes in RNA-Seq data.
• Performance variations suggest challenges in combining p-values, potentially due to test dependency.
• A modified Fisher's method may offer improved accuracy when combining p-values, especially with potential dependencies.