We have developed another approximation that also constrains phasing updates to condition on a subset of k haplotypes. Rather than selecting haplotypes at random, our approach seeks to identify the k haplotypes that are in some sense “closest” to the haplotypes of the individual being updated. In genealogical terms, this amounts to focusing attention on the parts of the underlying tree where that individual's haplotypes are located. The idea is that haplotypes that reside nearby in the genealogical tree will the most informative about the haplotypes of interest.
