A dictionary model for haplotyping, genotype calling, and association testing
Article first published online: 8 MAY 2007
© 2007 Wiley-Liss, Inc.
Volume 31, Issue 7, pages 672–683, November 2007
How to Cite
Ayers, K. L., Sabatti, C. and Lange, K. (2007), A dictionary model for haplotyping, genotype calling, and association testing. Genet. Epidemiol., 31: 672–683. doi: 10.1002/gepi.20232
- Issue published online: 16 OCT 2007
- Article first published online: 8 MAY 2007
- Manuscript Accepted: 25 MAR 2007
- Manuscript Revised: 20 FEB 2007
- Manuscript Received: 26 SEP 2006
- NIH training grant. Grant Number: HG02536
- NSF grant. Grant Number: DMS0239427
- USPHS grants. Grant Number: GM53275 and MH59490.
- linkage disequilibrium;
- forward algorithm;
- backward algorithm;
- Gibbs sampling
We propose a new method for haplotyping, genotype calling, and association testing based on a dictionary model for haplotypes. In this framework, a haplotype arises as a concatenation of conserved haplotype segments, drawn from a predefined dictionary according to segment specific probabilities. The observed data consist of unphased multimarker genotypes gathered on a random sample of unrelated individuals. These genotypes are subject to mutation, genotyping errors, and missing data. The true pair of haplotypes corresponding to a person's multimarker genotype is reconstructed using a Markov chain that visits haplotype pairs according to their posterior probabilities. Our implementation of the chain alternates Gibbs steps, which rearrange the phase of a single marker, and Metropolis steps, which swap maternal and paternal haplotypes from a given maker onward. Output of the chain include the most likely haplotype pairs, the most likely genotypes at each marker, and the expected number of occurrences of each haplotype segment. Reconstruction accuracy is comparable to that achieved by the best existing algorithms. More importantly, the dictionary model yields expected counts of conserved haplotype segments. These imputed counts can serve as genetic predictors in association studies, as we illustrate by examples on cystic fibrosis, Friedreich's ataxia, and angiotensin-I converting enzyme levels. Genet. Epidemiol. © 2007 Wiley-Liss, Inc.