A comprehensive comparison of multilocus association methods with summary statistics in genome-wide association studies

Abstract

Background Multilocus analysis on a set of single nucleotide polymorphisms (SNPs) pre-assigned within a gene constitutes a valuable complement to single-marker analysis by aggregating data on complex traits in a biologically meaningful way. However, despite the existence of a wide variety of SNP-set methods, few comprehensive comparison studies have been previously performed to evaluate the effectiveness of these methods. Results We herein sought to fill this knowledge gap by conducting a comprehensive empirical comparison for 22 commonly-used summary-statistics based SNP-set methods. We showed that only seven methods could effectively control the type I error, and that these well-calibrated approaches had varying power performance under the simulation scenarios. Overall, we confirmed that the burden test was generally underpowered and score-based variance component tests (e.g., sequence kernel association test) were much powerful under the polygenic genetic architecture in both common and rare variant association analyses. We further revealed that two linkage-disequilibrium-free P value combination methods (e.g., harmonic mean P value method and aggregated Cauchy association test) behaved very well under the sparse genetic architecture in simulations and real-data applications to common and rare variant association analyses as well as in expression quantitative trait loci weighted integrative analysis. We also assessed the scalability of these approaches by recording computational time and found that all these methods can be scalable to biobank-scale data although some might be relatively slow. Conclusion In conclusion, we hope that our findings can offer an important guidance on how to choose appropriate multilocus association analysis methods in post-GWAS era. All the SNP-set methods are implemented in the R package called MCA, which is freely available at https://github.com/biostatpzeng/.

Detecting associated genes for complex traits shared across East Asian and European populations under the framework of composite null hypothesis testing

“…Intuitively, the performance of a gene-set association analysis method is determined by how well its modeling assumption matches the true genetic architecture of a group of SNPs [83,84], which however is generally unknown a priori and varies from one gene to another. Consequently, it is difficult to choose a consistently optimal gene-set based method for all genes across the whole genome [52,85,86]. In the present study, we applied MAGMA to simultaneously examine the association evidence of multiple SNPs, which was in nature a variance-component based score test for multilocus genetic association studies built based on random-effects models [53,87].…”

“…In the present study, we applied MAGMA to simultaneously examine the association evidence of multiple SNPs, which was in nature a variance-component based score test for multilocus genetic association studies built based on random-effects models [53,87]. Although MAGMA might be not the optimal method for every gene, it exhibited excellent performance in statistical power compared to many existing gene-set based methods and was widely used in gene-centric integrative analysis in post-GWAS era [52][53][54][55].…”

“…These gene-level summary statistics can be easily to obtain with SNP-level association results publicly released by GWASs [48][49][50][51]. Therefore, we first aggregate multiple association signals at the SNP level into a single association signal at the gene level [52]. To this aim, we employ a powerful gene-set based association method called MAGMA [53], which is efficiently conducted via user-friendly software.…”