To explore the relative contributions of common versus rare causal variants to the genetic liability of AD, we simulated a series of disease models and conducted the same two-stage, genome-wide scoring for EA and AA samples, with routines delineated by MAF class. What we find is that the best fitting models, overall, are those based entirely on rare causal variants (Fig. 3). Although these simulations examined only a limited number of possible disease architectures and therefore do not preclude the possibility of thousands or tens of thousands of common loci solely contributing to AD risk, especially for heritabilities larger than the one tested in our models (65%), it does indicate that polymarker scoring based on GWAS data for complex phenotypes can detect the small, collective effects of rare and uncommon genetic determinants and that there could be as few as one hundred of them. This contrasts with the conclusion reached by Purcell et al. (2009) in their models that simulated both disease status and genotype data, asserting that rare variants could not alone account for R2 signals generated from genome-wide,
