Department of Statistics Seminar
North Carolina State University

presents

S. Stanley Young

Glaxo Wellcome Inc.

"The Discovery of Multiple, Interacting Genes"

(Joint Work with Andrew Rusinko, Glaxo Wellcome Inc., and Dimitri Zaykin, North Carolina State University)

ABSTRACT

Many of the disease traits of individuals are likely to be under the control of multiple, interacting genes. With the advent of DNA chips, multiple genetic markers and large scale experiments, data will become available for the discovery of multiple gene effects. But the problems are formidable: The logistics of data collection, cleaning, and storage will be difficult, time consuming, and expensive; we may only be able to observe marker genes linked to disease genes; there is no current statistical analysis method for linking phenotype to complex combinations of the 50 to 100 thousand human genes; the individual gene effects may be small; there is a mixture problem as a phenotype can be the result of different genotypes, i.e. there may be multiple mechanisms that lead to the same named disease. Recursive partitioning, RP, is a statistical technique for finding complex interactions among many variables to explain a response. RP easily handles large numbers of cases and mixtures; we extend RP to handle very large numbers of variables, tens of thousands to millions. Attention is paid to algorithmic speed, computer storage, and the multiple testing problem. Our methods are explicated with a molecular structure-biological activity data set; we then show how these methods can be applied to establish a genotype-phenotype relationship, GPR. The benefits of this method are many-fold, but include the association of phenotype to multiple, possibly interacting, gene effects.

Friday, October 17, 1997

3:35 - 4:35 pm

206 Cox Hall

Refreshments will be served on the second floor of Dabney Hall (left of Room 222) at 3:00 pm.