and estimates π0, defined as the proportion of hypotheses for which the null is true. Implicitly, the BH procedure assumes π0 is 1, whereas the ST procedure learns it from the data, resulting in more statistically significant hits. Of note, the adjusted P-values provided by FastQTL allow the users to easily apply whichever multiple testing correction they favor, from FDR to Bonferroni, since it provides adjusted P-values well calibrated on the full P-value range.
