A Two-Layer Hidden Markov Model for Detecting Loss of Heterozygosity

• Activity Number: 4
• Type: Invited
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Paul Scheet*+ and Selina Vattathil and Yongtao Guan
• Companies: The University of Texas MD Anderson Cancer Center and The University of Texas MD Anderson Cancer Center and Baylor College of Medicine
• Keywords: statistical genetics ; loss of heterozygosity ; hidden markov models

Abstract

In SNP microarray data from unpaired tumor samples or constitutional DNA only, inferring loss of heterozygosity (LOH) is typically conducted based on patterns of SNP genotypes, namely the existence of unusually long runs of homozygosity. However, run length alone does not take into account all available information, since local variation in allele and haplotype frequencies also affects the size of homozygous regions. Here we present a model that accounts for these features of the data to improve inference of LOH due to somatic events. The statistical model is based on a 2-layer hidden Markov model (HMM). In traditional "single layer" HMMs for genotype data, the hidden states represent template or consensus haplotypes. In this model, the hidden state space is expanded to include an LOH process that, when active, renders alleles from a single haplotype to be presented as homozygous genotypes. We apply our method to simulated data and compare results to those from more restrictive implementations to test the utility gained from relaxing individual assumptions, such as allowing for varying allele frequencies, accounting for the dependence among alleles at nearby markers.