Learning from our GWAS mistakes: from experimental design to scientific method.

Consider, though, that when genome-wide significance is not reached, researchers commonly take forward, say, 20–40 nominally significant signals from a GWAS and then run association tests for those signals in a second study, concluding that all the signals with a p -value <0.05 have replicated (no Bonferroni adjustment).

also in 6,706 other papers