We present FastQTL, a QTL mapper in cis for molecular phenotypes that implements a new permutation scheme to accurately and rapidly correct for multiple-testing at both the genotype and phenotype levels. FastQTL has several advantages compared with existing methods, making it the ideal candidate to map QTLs for the coming wave of large-scale datasets regrouping many different layers of molecular phenotypes and near complete collection of variant sites. First, permutations are modeled with a beta distribution, parameterized from a relatively small number of permutations. This results in accurate adjusted P-values which could not be feasibly obtained by standard or adaptive permutation analysis; for example down to 10−128 in the Geuvadis dataset. In practice, having well-calibrated adjusted P-values on the full range (0, 1) is of crucial importance to (1) estimate the number of tests made under the null (quantity underlying efficient FDR correction methods) and to (2) meta-analyze multiple QTL studies together. Second, the beta approximation behaves well enough with only 100 permutations to rapidly assess the impact on the analysis of important parameters such as cis-window size and covariates
