In hypothesis testing, p-value is
routinely used as a measure of statistical evidence against the null
hypothesis, where a smaller p-value indicates stronger evidence substantiating
the alternative hypothesis. P-value is the probability of type-I error made in a hypothesis testing, namely, the chance that one falsely reject the null
hypothesis when the null holds true. In a disease genome wide association study
(GWAS), p-value potentially tells us how likely a putative disease associated
variant is due to random chance. For a long time p-values have been taken
seriously by the GWAS community as a safeguard against false positives.
Every
disease-associated mutation reported in a GWAS must reach a stringent p-value
cutoff in order to survive the multiple testing corrections. This is reasonable because after testing millions of variants in the genome, some random variants ought to yield small p-values purely by chance. Despite of
p-value’s theoretical justification, however, it has become increasingly
evident that statistical p-values are not nearly as reliable as it was
believed.
