Churchill 9706863 The investigator develops new statistical methods for problems in molecular biology. Three problem areas are addressed: sequence alignment, phylogeny construction and genetic mapping. Although diverse and individually well developed, these problems can be unified under a common statistical framework. In each problem, a graph-valued random variable is of primary interest. Hierarchical Bayesian models are developed for each problem with an emphasis on common features. The techniques used to compute posterior probabilities from these models are Markov chain Monte Carlo methods. This computational approach is neccessary because analytic solutions for these problems are not attainable. The overall goal of this project is to develop an inference framework that is of general utility in a variety of problems. Toward this end, simple models are considered first. The simple models are generalized to the extent that it is feasible. This project does not solve all of the problems of computational molecular biology. The purpose of investigating a broad range of problems is to develop a common framework for statistical inference. This project develops new methods for the analysis of data that arise in molecular biology and genetics. These statistical methods have a wide variety of applications and are motivated by the dramatic increase in molecular data due to the expansion of biotechnology and the human genome project. Ultimately this leads to a deeper understanding of evolution and the relationships among different groups of organisms and their individual genes. It also addresses the problem of locating the genes that affect specific traits in humans, plants and animals. The greatest potential impact is on gene mapping in agricultural plants, due to ongoing collaborations with plant breeders at Cornell University. Although the project emphasizes statistical theory underlying the analysis of data, the methods are implemented in softw are that is useful to other researchers.
