Haplotype Inference using Expectation-Maximization: First, Find the Right Hill

• Activity Number: 283
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Benjamin Salisbury*+ and Andreas Windemuth and J. Claiborne Stephens
• Companies: Genaissance Pharmaceuticals, Inc. and Genaissance Pharmaceuticals, Inc. and Genaissance Pharmaceuticals, Inc.
• Keywords: haplotype ; expectation-maximization ; genomics

Abstract

A growing interest in genomics is the discovery of genetic differences that explain phenotypic differences between individuals. To discover these associations, we may often need to know how the alleles of atomistic units of variation such as SNPs and STRs co-occur on chromosomes. These assemblages are called haplotypes. Because it is difficult and expensive to determine haplotypes in the lab, statistics is often brought to bear for phasing the points of variation into the two chromosomal haplotypes that people and other diploid organisms have for most genes.

We have studied empirically, using hundreds of natural test cases, how Expectation-Maximization (EM) can be helpful in inferring haplotypes from unphased genotypes. EM is a useful tool for climbing hills on a likelihood landscape. However, this leaves the question of which hill(s) to climb. We compare initialization approaches that have been used by others and some that we have developed. By testing the results from each choice against pedigree-based deductions, we can confidently conclude that some approaches are far more successful than others at recovering correct haplotypes.