Abstract :
[en] Genome-wide association interaction (GWAI) studies have increased in popularity. Yet
to date, no standard protocol exists. In practice, any GWAI workflow involves making
choices about quality control strategy, SNP filtering, linkage disequilibrium (LD)
pruning, analytic tool to model or to test for genetic interactions. Each of these can
have an impact on the final epistasis findings and may affect their reproducibility in
follow-up analyses. Choosing an analytic tool is not straightforward, as different such
tools exist and current understanding about their performance is based on often very
particular simulation settings. In the present study, we wish to create awareness for the
impact of (minor) changes in a GWAI analysis protocol can have on final epistasis
findings. In particular, we investigate the influence of marker selection and marker
prioritization strategies, LD pruning and the choice of epistasis detection analytics on
study results, giving rise to 8 GWAI protocols. Discussions are made in the context of
the ankylosing spondylitis (AS) data obtained via the Wellcome Trust Case Control
Consortium (WTCCC2). As expected, the largest impact on AS epistasis findings is
caused by the choice of marker selection criterion, followed by marker coding and LD
pruning. In MB-MDR, co-dominant coding of main effects is more robust to the effects
of LD pruning than additive coding. We were able to reproduce previously reported
epistasis involvement of HLA-B and ERAP1 in AS pathology. In addition, our results
suggest involvement of MAGI3 and PARK2, responsible for cell adhesion and cellular
trafficking. Gene Ontology (GO) biological function enrichment analysis across the 8
considered GWAI protocols also suggested that AS could be associated to the Central
Nervous System (CNS) malfunctions, specifically, in nerve impulse propagation and in
neurotransmitters metabolic processes.
Scopus citations®
without self-citations
6
