Multi-Layer Clustering and Its Applications: Local Ancestry Inference and Haplotype-Based Association Mapping

• Activity Number: 588
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yongtao Guan*+
• Companies: Baylor College of Medicine
• Keywords: local ancestry ; admixture mapping ; haplotype association ; hidden Markov model

Abstract

We present a multi-layer clustering model that aims to emulate a coalescent process with recombination, where a marginal coalescent tree is approximated by a hierarchy of star trees, and ancestral recombination is approximated by switch of clusters within each hierarchy. In the simplest form of a two layer clustering model, depending on the prior for the cluster-switching frequencies, the bottom layer clustering may represent fine haplotypes and the top layer clustering may represent the structure of such haplotypes, either due to the difference in ancestry or due to subtle difference in haplotype frequencies in case/control status. In the former scenario, one can accurately infer local ancestries for admixed individuals; and in the later scenario, one can detect haplotype based genotype-phenotype association. The model is easy to fit with standard hidden Markov model techniques, can accommodate more than two source populations, and does not rely on training data with known haplotypes.