Association Analyses
For each protein and transcript trait, interindividual levels were inverse normal transformed and tested for association with all markers genome-wide. Association testing was performed by linear regression implemented in Python and R using custom scripts. For each trait, we selected the most significantly associated SNV within each recombination window, defined by splitting the genome into 25,307 blocks flanked by >10 cM/Mb recombination rates estimated from HapMap.22 All SNV-protein associations with p < 10−4 for proteins with more than one biological replicate were validated with the linear mixed-effects model y ∼p + G + T|I + e with a fixed genotype variable, G.