Then, we looked at the downstream impact of the beta approximation on QTL discovery. To do so, we first generated an optimal 5% FDR eQTL set for GEUV_EUR by running 100 000 permutations and then measured the sensitivity/specificity ratios of reasonable FastQTL configurations to recover this optimal set. Specifically, we run from 50 to 5000 permutations using either the beta approximation or the direct method to compute adjusted P-values. We find that using 500–5000 permutations allows us to approximate well the optimal set (Fig. 2d).
