MACH implements a Markov chain Monte Carlo-based algorithm to infer possible pairs of haplotypes for each individual's genotypes (including untyped genotypes) [Li et al., 2006; Li et al., 2007]. Following the authors’ recommendation in the provided documentation, we used a two-stage procedure. First, we inferred haplotype phase for the reference panel using 50 rounds of the Markov chain sampler and allowing at most 200 haplotypes when updating the phase for each individual. Second, we conditioned imputation on the first-stage maximum-likelihood estimates of the crossover map, which specifies the likely locations of haplotype transitions, and the error rate map, which specifies unusual markers based on a combination of discrepancies between the reference panel and study sample data, genotyping error, and recurrent mutation. MACH produces an output file containing the posterior probabilities for two of the three genotypes at each SNP for each individual. Software is available at http://www.sph.umich.edu/csg/abecasis/MACH/download/.
