BEAGLE uses a localized haplotype clustering-based algorithm [Browning and Browning, 2007]. First, it clusters haplotypes at each marker and defines a hidden Markov model to find the most likely haplotype pairs based on the individual's known genotypes. Then, the most likely genotype at untyped loci can be imputed from final haplotype pairs. Due to extensive memory usage, we cut the study sample into three sets of 500 individuals and one set of 476 individuals and we invoked the low memory command line option. By default, the posterior probabilities for the three genotypes at each SNP for each individual are printed in an output file. Software is available at http://www.stat.auckland.ac.nz/~browning/beagle/beagle.html.
