We used the following methods to perform the imputation: IMPUTE v1.0; MACH v0.1.10 with analytical (“mle”) imputation, where the model parameters were selected by running the “greedy” algorithm for 100 iterations on a random subset of 500 58 C samples, as suggested in the online tutorial that accompanies the software (http://www.sph.umich.edu/csg/abecasis/mach/tour/imputation.html); fastPHASE [11] v1.3.2 with 20 and 30 clusters (K = 20 and K = 30, in separate runs), 15 starts of the expectation-maximization (EM) algorithm to estimate model parameters, and 35 iterations per EM start (this version of fastPHASE automatically fits the clustering model to the reference panel and then imputes each study individual separately, conditional on the fitted model); BEAGLE v3.0.2 on default settings and with 50 iterations (rather than the default 10); and IMPUTE v2.0 with 40 and 80 conditioning states used for diploid updates at typed SNPs (k = 40 and k = 80, in separate runs) and 120 conditioning states (i.e., the full HapMap CEU panel) used for all haploid updates. We ran IMPUTE v2 with 10 burn-in iterations followed by 20 additional iterations. The
