Multiple testing on the directed acyclic graph of gene ontology

Area of Science:

• Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Existing multiplicity adjustment methods do not utilize the Gene Ontology (GO) graph structure.
• Testing for associations between gene expression profiles of GO terms and a response variable is a common task.

Purpose of the Study:

• To propose a novel multiple testing procedure that incorporates the GO graph structure.
• To develop a method that controls the family-wise error rate (FWER) while considering the hierarchical nature of GO terms.

Main Methods:

• The focus level procedure combines a Closed Testing procedure with Holm's method.
• It requires users to specify a 'focus level' in the GO graph, indicating their preferred specificity.
• An algorithm for calculating multiplicity-adjusted P-values is provided.

Main Results:

• The focus level procedure strongly controls the family-wise error rate (FWER) without assumptions on test statistic distributions.
• It enhances power at the user-defined focus level within the GO graph.
• Adjusted P-values may not preserve the original ordering of raw P-values due to GO graph structure preservation.

Conclusions:

• The focus level procedure offers a powerful and flexible approach for multiplicity adjustment in GO-based gene expression analysis.
• This method effectively utilizes the GO graph structure for improved statistical inference.
• The procedure is implemented in the globaltest and GlobalAncova R packages.