Feasible and Successful: Genome-Wide Interaction Analysis Involving All 1.9 × 10¹¹ Pair-Wise Interaction Tests

Abstract

The Genome-Wide Association Study (GWAS) is the study design of choice for detecting common genetic risk factors for multifactorial diseases. The performance of full Genome-Wide Interaction Analyses (GWIA) has always been considered computationally challenging. Two-stage strategies to reduce the amount of numerical analysis require the detection of single marker effects or prior pathophysiological hypotheses before the analysis of interaction. This prevents the detection of pure epistatic effects. Our case-control study in idiopathic generalized epilepsy demonstrates that a full GWIA is feasible through use of data compression, specific data representation, interleaved data organization, and parallelization of the analysis on a multi-processor system. Following extensive quality control of the genotypes, our final list of top interaction hits contains only pairs of interacting SNPs with negligible marginal effects. The TOP HIT interaction was between a SNP-pair intragenic to gene DNER (chr 2) and gene CTNNA3 (chr 10). Both of these genes are functionally involved in neuronal migration, synaptogenesis, and the formation of neuronal circuits. Our results therefore indicate a possible interaction between these two genes in epileptogenesis. Results from GWAS are beginning to reveal a ‘missing heritability’ in complex traits and diseases. Systematic, hypothesis-free analysis of epistatic interaction (GWIA) may help to close this increasingly recognized gap in heritability.

Significance Levels in Genome-Wide Interaction Analysis (GWIA)

“…As for single-marker analysis, we believe that it is important to rely on a commonly accepted significance cut-off for GWIA studies. In addition, it is obligatory to follow the principles of the NCI-NHGRI Working Group on Replication in Association Studies (2007), since the multiple testing in GWIA is immense and interaction tests are even more seriously affected by undetected genotyping errors than are single-marker tests (Steffens et al, 2010). The multiple testing penalty we computed for GWIA is higher than we ourselves had expected and raises the question whether GWAS data of typical size provide enough power to detect interaction.…”

“…With 500,000 SNPs, 1.25 × 10 11 SNP pairs have to be tested for interaction. While we have recently conducted a complete GWIA using 256 parallel computing nodes (Steffens et al, 2010), we estimated a running time of 80 days for a complete GWIA of these SNPs with 1200 individuals on a single 3 GHz computer, even with a particularly fast test for interaction (see methods). As a consequence, conduction of type I error or power studies based on thousands of simulated genome-wide data sets is impossible.…”

Investigation of tryptophan hydroxylase 2 (TPH2) in schizophrenia and in the response to antipsychotics

“…On the other hand, it is also possible that a small number of genotyping errors at one of the SNPs have led to an artifact. Such errors are typically not detectable by the application of QC criteria that focus on single SNPs (Steffens et al, 2010).…”

A systematic eQTL study of cis–trans epistasis in 210 HapMap individuals

“…Second, systematic two-locus eQTL mappings require substantial computational resources, although this limitation has recently been overcome by the introduction of novel biostatistical methods. [17][18][19] In the present study we tried to circumvent some of the limitations associated to interaction scans and performed a systematic two-locus eQTL study for epistasis. Out of three possible two-locus interaction models (ie, cis-cis, cis-trans, trans-trans), we restricted our analysis only to cis-trans epistasis.…”