An investigation of two multivariate permutation methods for controlling the false discovery proportion

Area of Science:

• Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• Microarray experiments aim to identify differentially expressed genes.
• Balancing the discovery of true positives against false discoveries is a key challenge.
• Controlling the false discovery proportion (FDP) is essential for reliable gene expression analysis.

Purpose of the Study:

• To investigate two multivariate permutation methods for controlling the FDP in gene expression studies.
• To compare the performance of Multivariate Permutation Testing (MPT) and Significance Analysis of Microarrays (SAM) in FDP control.
• To assess the impact of implementation strategies (top-down vs. bottom-up) on FDP control.

Main Methods:

• Evaluation of a Multivariate Permutation Testing (MPT) method for probabilistic FDP control.
• Assessment of the Significance Analysis of Microarrays (SAM) procedure for FDP estimation.
• Comparison of 'top-down' and 'bottom-up' approaches for both MPT and SAM.
• Consideration of gene correlations within the multivariate methods.

Main Results:

• The effectiveness of MPT and SAM in controlling FDP varied significantly based on implementation.
• The 'top-down' MPT-based method probabilistically controlled the FDP.
• The 'top-down' SAM-based method, as implemented, did not adequately control the FDP.
• Both 'bottom-up' MPT and SAM methods demonstrated poor FDP control.

Conclusions:

• The 'top-down' MPT approach offers a reliable strategy for controlling the false discovery proportion in gene expression analysis.
• Implementation details critically influence the performance of permutation-based methods for FDP control.
• Careful selection and implementation of statistical methods are necessary to minimize false discoveries in high-throughput genomic studies.