Using Markov Chain Composite Likelihood to Analyze Long Sequence Data

• Activity Number: 301
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Jianping Sun*+ and Bruce George Lindsay
• Companies: Fred Hutchinson Cancer Research Center and Penn State University
• Keywords: Markov Chain ; Composite Likelihood ; Long Sequence Data ; Mutation ; Recombination

Abstract

The primary goal of this talk is the analysis of long sequence data generated in biology, such as SNP data. Suppose we have observed n current descendant sequences of length L, one interesting question is that how to estimate the unknown ancestral distribution from the observed descendants, considering realistic biology complexities such as mutation and recombination. We have developed a statistical model with both mutation and recombination to estimate the ancestral distribution. However, though we can write out the full likelihood for ancestral distribution explicitly, there is an enormous computation challenge when applying it on data due to an enormous number of recombination possibilities, which grows exponentially in sequence length. Therefore, we apply composite likelihood as an approximation to solve the problem.

In this talk, we first introduce our developed statistical model and composite likelihood method. Then, a Markov chain composite likelihood (MCCL) method is proposed and applied to our statistical model. Finally, some simulation results are shown to investigate the performance of the MCCL.