A comparison of statistical tests for detecting differential expression using Affymetrix oligonucleotide microarrays

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate signal quantification and differential expression detection are crucial for analyzing Affymetrix microarray data.
• Numerous methods exist, necessitating comparative evaluation for optimal performance.

Purpose of the Study:

• To evaluate and compare various signal quantification methods (GCRMA, RSVD, VSN, MAS5, Resolver) and statistical methods for differential expression (t test, Cyber-T, SAM, LPE, RankProducts, Resolver RatioBuild).
• To assess the effectiveness of these methods in detecting differential gene expression using two distinct datasets.

Main Methods:

• Comparative analysis of signal quantification algorithms: GCRMA, RSVD, VSN, MAS5, and Resolver.
• Evaluation of differential expression detection methods: t test, Cyber-T, SAM, LPE, RankProducts, and Resolver RatioBuild.
• Utilized two datasets: Affymetrix HG-U133 Latin Square spike-in and an in-house rat liver transcriptomics study.

Main Results:

• GCRMA is recommended as the superior method for signal quantification.
• GCRMA combined with Cyber-T or SAM demonstrated the highest performance in detecting differential expression, as measured by the area under the ROC curve.
• Resolver RatioBuild provides a competitive integrated alternative for both quantification and differential expression analysis.
• MAS5 signal quantification generally resulted in inferior performance for most differential expression algorithms.

Conclusions:

• GCRMA is the preferred method for signal quantification in Affymetrix microarray analysis.
• The combination of GCRMA with Cyber-T or SAM is the most effective strategy for detecting differential gene expression.
• Resolver RatioBuild offers a robust integrated pipeline for comprehensive analysis.