Abstract:
|
In genetic association studies, multidimensional correlated mixed responses with a large number of covariates, such as SNPs, are frequently collected, but many of these covariates may be irrelevant or even noises to prediction problem. For the reason of generalization performance of prediction and scientific considerations, one would like to select a subset of relevant features to improve the ability of genetic marker discovery and prediction. We propose a joint penalized approach for correlated mixed responses with application to feature selection. We develop computationally feasible computing algorithms to fit the proposed models. Numerical results demonstrate that the percentage of crucial genetic predictors selected by the proposed approach is much higher than marginal regression analysis. An example of real genetic data analysis is provided to examine the performance of the proposed approach.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.