The recent popularity of penalized regression in high-dimensional dataanalysis has lead to a demand for new inferential tools for these models. Falsediscovery rate control is a well known aspect of other high-dimensional dataanalysis approaches that has only recently been considered in the context ofpenalized regression. However much of the current research has been focusedprimarily on lasso-penalized linear regression, despite the availability ofsoftware that can fit penalized regression models to a variety oflikelihood-based models. In this paper we derive a method of estimating themarginal false discovery rate for penalized likelihood methods and demonstrateits application to penalized logistic and penalized Cox regression models. Ourapproach is fast, flexible and can be applied to a variety of penalty functionsincluding lasso, elastic net, MCP, and MNet. We derive theoretical resultsunder which the proposed estimator is valid, and use simulation studies todemonstrate that the approach is reasonably robust, albeit slightlyconservative, when these assumptions are violated. The practical utility of themethod is demonstrated on two gene expression data sets with binary and surivaloutcomes, which we respectively analyze using penalized logistic and penalizedCox regression.
展开▼
