Genomics research of UNC biostatisticians featured in Newsweek


The work of a team of UNC biostatisticians is described in the July 21 Newsweek science column, which states that “in the Wild West that is genome research, statisticians are the new sheriffs in town.” http://www.newsweek.com/id/145865.  The research was presented at the August Joint Statistical Meetings in Denver by Dr. Fred Wright, Professor of Biostatistics at the UNC, Director of the Carolina Environmental Bioinformatics Center and a project co-PI of the Carolina Center for Computational Toxicology.  Dr. Wright plans to apply the techniques to his work in UNC’s two EPA-funded Centers.

In modern genomic association studies, researchers attempt to find the genetic causes of a disease by comparing diseased individuals to healthy individuals at hundreds of thousands of genetic markers.  When performing so many statistical tests, it is widely recognized that stringent thresholds must be used in order to avoid false positives.  However, until recently it was not well known that the magnitude of the significant gene effects tends to be inflated.  This phenomenon has been dubbed the “winner’s curse,” a term borrowed from auction settings in which successful buyers tend to overpay.

“In genetic epidemiology, the winner’s curse can have a dramatic effect, resulting in exaggerated risk estimates,” explains Dr. Wright.  “As a result, a gene’s impact may not be properly understood, and follow-up studies may be designed with too few patients.”  To handle this problem, Biostatistics Ph.D. student Arpita Ghosh, under the direction of Wright and Associate Professor Fei Zou, devised a simple approach to reduce the estimated effect size using a mathematical technique known as conditional likelihood analysis.  In other words, the approach reverses the winner’s curse.  The initial method was published in the May 2008 issue of the American Journal of Human Genetics, and the team’s follow-up work addresses key extensions, such as handling situations involving the interaction of genes and environmental factors.

“The beauty of this approach is its simplicity, and it can be applied to many other types of studies and fields of science beyond genomics,” says Ghosh.  Adds Dr. Zou, “You can even apply this approach after a research finding has been published, simply by performing modified calculations on published tables of results.” 

Wright plans to apply a variant of his group’s statistical techniques to overcome the winner’s curse in computational toxicology, highlighting the wide applicability of the approach.  More broadly, the Newsweek article highlights the critical role that statisticians now play in all genomics research, including the techniques of modern toxicogenomics.

###

Research supported in part by the Carolina Environmental Bioinformatics Center, the Carolina Center for Computational Toxicology (U.S.E.P.A.) and the N.I.H.


 July 21, 2008