Combining Multiple Hypothesis Testing with Machine Learning Increases the Statistical Power of Genome-wide...

Area of Science:

• Genetics and Bioinformatics
• Statistical Genomics
• Computational Biology

Background:

• Genome-Wide Association Studies (GWAS) typically analyze single nucleotide polymorphisms (SNPs) individually.
• This standard approach may miss complex associations due to linkage disequilibrium and requires improvement for higher accuracy.

Purpose of the Study:

• To introduce COMBI, a novel two-step algorithm integrating machine learning and statistical testing for enhanced GWAS analysis.
• To improve the power and precision of SNP association detection in GWAS by accounting for correlation structures.

Main Methods:

• COMBI employs a two-step approach: first, a support vector machine (SVM) identifies candidate SNPs.
• Second, hypothesis tests are performed on candidate SNPs with appropriate threshold correction, considering SNP correlations.

Main Results:

• COMBI demonstrated superior performance compared to raw p-value thresholding and other state-of-the-art methods in validating against independent GWAS data.
• The algorithm achieved higher power and precision, with over 80% of its discoveries from WTCCC data being validated in later studies.

Conclusions:

• COMBI offers a more powerful and precise method for GWAS analysis, yielding more true and fewer false discoveries.
• The COMBI algorithm, available in the GWASpi toolbox, represents a significant advancement in genetic association studies.