Genotypes were prephased with SHAPEIT2 (61), and missing genotypes were imputed with IMPUTE2 (62) in 5-Mb chunks against the 1000 Genomes reference panel (63). Summary statistics for each chunk were obtained using SNPTEST (https://mathgen.stats.ox.ac.uk/genetics_software/snptest/snptest.html), and sites with an information score of less than 0.9 or significant departure from Hardy-Weinberg equilibrium (P < 1 × 10−3) were excluded from further analysis. Genotype probabilities for all remaining sites were converted into dosage estimates and formatted for use with MatrixEQTL.
