PubMed search for Dr. Michael P. Epstein

Areas of Specialization/Research Interests:
Statistical methods for gene mapping
Genetic mapping of Mendelian and complex traits

Professional Memberships and Activities:
American Society of Human Genetics
International Genetic Epidemiology Society

Education:
Ph.D., Biostatistics, University of Michigan, 2002
M.S., Biostatistics, University of Michigan, 1998
B.S., Mathematics and Biological Anthropology, Duke University, 1996

Research Interest Description:
My general area of research interest is statistical genetics. I am particularly interested in statistical methods for human gene mapping of qualitative and quantitative hereditary traits. Technological advances in modern molecular genetics as well as the extraordinary amount of information being generated by the Human Genome Project make this an especially exciting area of research. Identifying the genetic determinants of hereditary traits will require the development of sophisticated statistical and computational tools for the foreseeable future.

My current methodological work focuses on variance-component linkage methods for non-normally distributed trait data. Variance-component methods of quantitative trait linkage analysis have many attractive features that other quantitative trait linkage methods lack. Variance-component procedures require few assumptions and can accommodate many genetic phenomena including multiple gene and environmental effects and their interactions. Variance-component approaches also tend to have greater linkage power than relative-pair based techniques, in part because variance-component methods analyze all individuals in a family simultaneously.

However, standard variance-component methods assume that the analyzed quantitative trait data in a family follows a multivariate normal distribution. If this assumption is violated, variance-component methods may yield biased results. This limits the application of these methods. Therefore, I am currently extending the variance-components approach of linkage analysis to accommodate non-normally distributed trait data. I still use the general variance-component modeling framework, but apply generalized linear mixed models to account for the non-normal nature of the data. My present work focuses on a tobit variance-component method for trait data that are truncated due to assay limitation or medication effects. Future work will apply these generalized linear mixed modeling procedures further to perform variance-component linkage analyses on other types of non-normally distributed trait data that are collected in a genetic study. I plan on developing a logistic variance-component method to analyze binary data (e.g. disease status), a proportional-odds variance-component method to analyze polychotomous data, and a log-linear variance-component method to analyze count data. Linkage analyses using these novel variance-component methods should increase our ability to map genes influencing non-normally distributed trait data.

My future work also will extend the generalized linear mixed modeling procedures to perform family-based tests of allelic association of a marker with a gene influencing a non-normally distributed trait (such as disease). Allelic association tests often are a necessary approach for mapping trait-influencing genes, particularly after linkage studies specify a region of interest for future analysis. Such association tests look for marker alleles that are in linkage disequilibrium (association in the presence of linkage) with a nearby trait-influencing gene. Linkage disequilibrium occurs when the marker is very close to the gene such that the alleles of the marker and gene are transmitted together through multiple generations. Therefore, tests for linkage disequilibrium help fine-map the location of the gene. My planned family-based test is based on an idea that Abecasis et al. (2000) used for testing linkage disequilibrium with a gene that influenced a normally distributed trait. Besides having the flexibility to adjust for multiple genetic and non-genetic effects, my proposed test for linkage disequilibrium also controls for population stratification that would otherwise lead to an increase in false-positive findings.