Moment based gene set tests.

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Permutation-based gene set tests are standard for high-throughput expression analyses.
• Current methods require a large number of permutations (M) for smaller p-values, increasing computational cost.
• Parametric approximations to permutation distributions are sought to reduce this cost.

Purpose of the Study:

• To develop and evaluate a moment-based parametric approximation for gene set test statistics.
• To reduce the computational cost of gene set enrichment analysis.
• To enable faster identification of significant gene sets in large biological datasets.

Main Methods:

• Calculated exact moments of sum and sum of squared correlation statistics.
• Fitted parametric distributions to these moments.
• Applied the method to Parkinson's Disease expression datasets.

Main Results:

• The moment-based approximation significantly reduces computational cost (e.g., |G| or |G|^2 permutations vs. M).
• Approximate p-values closely matched permutation method p-values.
• Identified novel enriched gene sets in Parkinson's Disease datasets.
• Rankings of gene sets were nearly identical between methods.

Conclusions:

• Developed a moment-based approximation for linear and quadratic gene set test statistics.
• Achieved orders of magnitude speedup compared to permutation sampling.
• Implemented the method as a publicly available Bioconductor package, npGSEA.