For simplicity, we will focus on a single molecular phenotype P quantified in a set of N samples. Let G be the set of genotype dosages at L variant sites located within a cis-window of ± W Mb of the genomic location of P. To discover the best candidate QTL for P, FastQTL measures Pearson product-moment correlation coefficients between P and all L variants in G, stores the most strongly correlated variant q ∈ G as candidate QTL, and assesses its significance by calculating a nominal P-value pn with standard significance tests for Pearson correlation. Note that this is equivalent to testing for β ≠ 0 in a linear model P = βg + ϵ with β estimated by least squares fitting. Of note, this method is also used by Matrix eQTL to speed up linear regression (Methods 3.1 & 3.2 of Shabalin, 2012). Then, two multiple-testing levels are accounted for to determine the whole-genome significance of this nominal P-value and thereby to consider the corresponding variant q as a QTL: multiple genetic variants are tested per phenotype and multiple
