A Hidden Markov Technique for Haplotype Reconstruction

We give a new algorithm for the genotype phasing problem. Our solution is based on a hidden Markov model for haplotypes. The model has a uniform structure, unlike most solutions proposed so far that model recombinations using haplotype blocks. In our model, the haplotypes can be seen as a result of iterated recombinations applied on a few founder haplotypes. We find maximum likelihood model of this type by using the EM algorithm. We show how to solve the subtleties of the EM algorithm that arise when genotypes are generated using a haplotype model. We compare our method to the well-known currently available algorithms (phase, hap, gerbil) using some standard and new datasets. Our algorithm is relatively fast and gives results that are always best or second best among the methods compared.