I think our results, in line with other recent findings, have important implications for GWAS methodology. Complex systems are rarely well captured by linear models; it’s reasonable to expect the same in complex trait genomics. Feedback on the manuscript is most welcome.

We show that even if the variance explained by epistasis is modest, it's possible to get false hits. The problem is also positively correlated with sample size, suggesting caution when interpreting significant associations especially for large-scale GWAS.