Lastly, we employed the eQTL derived from the unpublished ROS/MAP study (https://www.synapse.org/#!Synapse:syn3219045, details on the study given below) in a limited way to replicate the 5 single-gene QTL associations we detected as having strong overlap with GWAS risk variants. The subset of the ROS/MAP cohort currently RNA-sequenced and analyzed (N = 461 DLPFC samples) was used to derive eQTL. To account for non-genetic factors such as batch effects, age, gender, and technical artifacts in the gene expression data, PEER70 was applied. The optimal numbers of PEER factors for association analysis were determined based on the factors that resulted in the maximal number of cis-eQTLs. This procedure identified between 30 and 40 factors in this DLFPC dataset; here, we used 30 PEER factors. We regressed out these factors from the gene expression levels and used the residuals as phenotypes for all eQTL association analyses. The ROS/MAP study 26 takes advantage of data and biological specimens from more than 1000 persons from two prospective, longitudinal clinical-pathologic studies of older subjects that are non-demented at the time of recruitment (the religious order study
