Home > Events > 2014 Seminars & Colloquia > KATHRYN ROEDER Carnegie Mellon University

KATHRYN ROEDER Carnegie Mellon University

Main Content

A method to exploit the structure of genetic ancestry space to enhance case-control studies
When
09 October 2014 from 4:00 PM to 5:00 PM
Where
201 Thomas Bldg.
Add event to calendar
vCal
iCal

In genetic studies of common and rare variants, considerable effort and expense is required to obtain a sample of control subjects matched by genetic ancestry to the case subjects.   Alternatively, repositories like dbGap already contain genetic data from tens of thousands of potential control samples.  These data can be accessed, but only with considerable effort on the part of the research team. It should be possible to model these data and obtain allele frequency estimates, which would obviate the need for collecting additional large control samples.  The task is challenging for two reasons: due to issues of privacy, genotype data can not be shared directly; and yet the control data must be chosen so that it is comparable in genetic ancestry to the particular case sample.  Our proposed approach, the Universal Control Repository Network (UNICORN), aims to provide allele frequency information that is optimally matched to the case sample.  To maintain the confidentiality of both cases and controls, no case genotype information is passed to UNICORN, nor will the controls available in the repository ever be accessible to external researchers. Instead we will use existing publicly available collections of control data to create a common genetic ancestry space onto which cases and control can be mapped independently. We use spectral clustering to construct ancestry spaces as well as to perform projections. The base space and projected controls are then used to estimate the allele frequency surface over the ancestry space. To identify small-scale frequency variation while also borrowing strength from the entire data set we employ a combination of empirical Bayesian analysis across a hierarchical clustering of the controls and, for localized ancestry regions, a Gaussian process model of the minor allele frequency.  We have performed a small-scale association test on the POPRES data based on simulated signal of varying risk and allele frequency and found that UNICORN delivers strongly superior results over a traditional matched control setup, even when the matched controls greatly outnumber cases. We believe that our proposed model will be of significant importance to researchers by enabling more powerful association studies with fewer resource expenditures.

Navigation for this Section

Events