5 (Ptraining). Alternatively, we also examined Ptraining thresholds with nonoverlapping ranges (10−5 < P < 10−4 to 0.4 < P < 0.5) to assess whether any of these finer threshold groups explained more variance in depression. These PS were calculated using PLINK's SNP scoring routine. The cross-validation procedure was repeated four times, rotating the testing set each time (a “leave-one-out” procedure). Genome-wide polygenic scoring from two external studies—GAIN-MDD (GAIN-MDD-PS) and PGC-MDD (PGC-MDD-PS) In addition, we attempted to replicate the published finding by Demirkan et al. (2011). Through
personal Inhibitors,research,lifescience,medical communication, Demirkan and colleagues provided the precise beta weights and P-values derived Inhibitors,research,lifescience,medical from their discovery set to facilitate replication in our cohorts. We also sought replication using data from another nine-study meta-analysis, which has been recently published (Ripke et al. 2013). We again considered the same nine P-value thresholds described above for selecting SNPs to be included in the PS calculation. The PS analysis was individually SCH772984 in vitro performed in each of the four NHS substudies and was Inhibitors,research,lifescience,medical meta-analyzed in the end. Candidate gene
polygenic scoring in NHS (candidate-PS) Some investigators have suggested that the candidate gene approach is less likely to yield true causal loci, with most positive results arising by chance. In that case, a candidate gene approach to PS may exacerbate Inhibitors,research,lifescience,medical the difficulty of such efforts. However, given the significant literature on candidate genes, the ongoing controversies regarding
which genes matter, and the substantial research attention Inhibitors,research,lifescience,medical they have received, the candidate gene polygenic scoring in NHS was also conducted and is described in detail in the Data S1. Briefly, to develop an informed candidate-PS, we selected 17 candidate genes with at least two positive prior reports of involvement in depression many on the PubMed via the HuGE Navigator (Yu et al. 2008) as of May 2011. Ultimately 96 independent SNPs were reserved for analysis, and each candidate gene was represented by at least one SNP. We used the same cross-validation procedure to obtain an unbiased estimate of the prediction performance as described above. Analyses of associations between PS and depression phenotypes Assessments of the association between PS and depression phenotype were performed in R with linear (for continuous outcome) or logistic (for dichotomized outcome) regressions. Covariates included in all PS analyses were the same as in the SNP GWAS analysis: age, case–control status in each original GWAS, and the three or four eigenvectors.