|
Abstract:
|
It is often of interest to find mediator variables which may partly explain the effect of a treatment on a response. Methods are well established for testing whether one or a few variables are significant mediators. However, if the mediators of interests are genes or measures of gene expression, there may be thousands of candidate mediators. This suggests the need to combine mediation analysis with high-dimensional data analysis techniques such as variable screening. Zhang and co-workers (2016, Bioinformatics) proposed a technique for studying high-dimensional mediation analysis. This involves screening and selection of potential mediators, on the basis of both their relationship with the treatment and also their relationship to the response after adjusting for the treatment. We illustrate their technique and some possible variations in an analysis of gene expression variables which may mediate the effect of the drug varenicline on reducing alcohol consumption in mice. We present possible candidate genes which may partly explain the drug's effect. We also present the results of a small simulation study on the performance of this approach.
|
