Abstract:
|
In modern pathway analysis, it is very common and biologically interesting to find genes that are common between multiple genetic pathways when multiple genetic pathways are functionally related. Therefore, tools and techniques are necessary which can take into account the overlapping features between the pathways for better pathway and gene selection purpose. In this paper, we develop a set of Bayesian Shrinkage models based on Bayesian lasso and Bayesian elastic-net, which can do variable selection with group information and overlapping features. The covariates that appear in multiple groups simultaneously creating the overlap among the groups are modeled by decomposing the effects of the overlapped variables into each of the membership pathway groups. The effectiveness of our Bayesian models are demonstrated by several multiple pathway analysis examples with publicly available real data sets and also with simulated data sets.
|