Article first published online: 3 FEB 2012
Copyright © 2012 John Wiley & Sons, Ltd.
Statistics in Medicine
Volume 31, Issue 10, pages 988–1000, 10 May 2012
How to Cite
Shahbaba, B., Shachaf, C. M. and Yu, Z. (2012), A pathway analysis method for genome-wide association studies. Statist. Med., 31: 988–1000. doi: 10.1002/sim.4477
- Issue published online: 11 APR 2012
- Article first published online: 3 FEB 2012
- Manuscript Accepted: 2 NOV 2011
- Manuscript Revised: 20 OCT 2011
- Manuscript Received: 22 MAR 2011
- Beaglecall. Z. Y.. Grant Number: NIH/R01 HG004960
- National Center for Research Resources (NCRR). Grant Number: UL1 RR031985
For genome-wide association studies, we propose a new method for identifying significant biological pathways. In this approach, we aggregate data across single-nucleotide polymorphisms to obtain summary measures at the gene level. We then use a hierarchical Bayesian model, which takes the gene-level summary measures as data, in order to evaluate the relevance of each pathway to an outcome of interest (e.g., disease status). Although shifting the focus of analysis from individual genes to pathways has proven to improve the statistical power and provide more robust results, such methods tend to eliminate a large number of genes whose pathways are unknown. For these genes, we propose to use a Bayesian multinomial logit model to predict the associated pathways by using the genes with known pathways as the training data. Our hierarchical Bayesian model takes the uncertainty regarding the pathway predictions into account while assessing the significance of pathways. We apply our method to two independent studies on type 2 diabetes and show that the overlap between the results from the two studies is statistically significant. We also evaluate our approach on the basis of simulated data. Copyright © 2012 John Wiley & Sons, Ltd.