AUTHOR=Marczyk Michal , Macioszek Agnieszka , Tobiasz Joanna , Polanska Joanna , Zyla Joanna TITLE=Importance of SNP Dependency Correction and Association Integration for Gene Set Analysis in Genome-Wide Association Studies JOURNAL=Frontiers in Genetics VOLUME=12 YEAR=2021 URL=https://www.frontiersin.org/journals/genetics/articles/10.3389/fgene.2021.767358 DOI=10.3389/fgene.2021.767358 ISSN=1664-8021 ABSTRACT=

A typical genome-wide association study (GWAS) analyzes millions of single-nucleotide polymorphisms (SNPs), several of which are in a region of the same gene. To conduct gene set analysis (GSA), information from SNPs needs to be unified at the gene level. A widely used practice is to use only the most relevant SNP per gene; however, there are other methods of integration that could be applied here. Also, the problem of nonrandom association of alleles at two or more loci is often neglected. Here, we tested the impact of incorporation of different integrations and linkage disequilibrium (LD) correction on the performance of several GSA methods. Matched normal and breast cancer samples from The Cancer Genome Atlas database were used to evaluate the performance of six GSA algorithms: Coincident Extreme Ranks in Numerical Observations (CERNO), Gene Set Enrichment Analysis (GSEA), GSEA-SNP, improved GSEA for GWAS (i-GSEA4GWAS), Meta-Analysis Gene-set Enrichment of variaNT Associations (MAGENTA), and Over-Representation Analysis (ORA). Association of SNPs to phenotype was calculated using modified McNemar’s test. Results for SNPs mapped to the same gene were integrated using Fisher and Stouffer methods and compared with the minimum p-value method. Four common measures were used to quantify the performance of all combinations of methods. Results of GSA analysis on GWAS were compared to the one performed on gene expression data. Comparing all evaluation metrics across different GSA algorithms, integrations, and LD correction, we highlighted CERNO, and MAGENTA with Stouffer as the most efficient. Applying LD correction increased prioritization and specificity of enrichment outcomes for all tested algorithms. When Fisher or Stouffer were used with LD, sensitivity and reproducibility were also better. Using any integration method was beneficial in comparison with a minimum p-value method in specific combinations. The correlation between GSA results from genomic and transcriptomic level was the highest when Stouffer integration was combined with LD correction. We thoroughly evaluated different approaches to GSA in GWAS in terms of performance to guide others to select the most effective combinations. We showed that LD correction and Stouffer integration could increase the performance of enrichment analysis and encourage the usage of these techniques.