Genome-wide association tests were conducted with the program PLINK (Purcell et al. 2007), using the standard measured genotype method, with covariates age and sex (quantile-quantile plots are provided in Fig. S4). SNPs were then delineated into bins according to incremental thresholds of association test P-values, as well as MAF ranges, from which scores were defined as the total number of “risk” alleles carried by a given target sample, weighted by the log odds ratio (OR) for AD as estimated from the COGA data. Scores were calculated for the SAGE data, limited to SNPs with an allele frequency > 1%, in HW equilibrium (P ≥ 1×10−4), and with a minimum call rate of 98% (n = 948,658). To measure how well the SAGE target scores predict AD risk, logistic regression analyses of case-control status were performed to quantify the amount of variation accounted for by the scores, as determined by Nagelkerke’s pseudo-R2, representing the difference in R2 estimates for the null model, with terms for the intercept, age, sex and genotyping rate, and the alternative model that includes the polymarker scores.
