Global Group Testing and Screening With Dynamic Effects

Identifying outcome-related variables is of general research interest in biomedical research. This task can be complicated by the presence of dynamic (or varying) variable effects that often manifest meaningful scientific mechanisms. Appropriately accounting for possible dynamic effects is crucial to avoid depreciating some important variables. In this work, we propose a model-free testing and screening framework by adopting a global view pertaining to the concept of interval quantile independence. The new framework not only permits robust identification of variables dynamically associated with an outcome, but also offers the flexibility to perform group testing that simultaneously evaluates multiple continuous or discrete covariates. We show that the key testing strategy can naturally evolve into unconditional and conditional screening procedures for ultra-high dimensional settings that enjoys the desirable sure screening property. We demonstrate good practical utility of the proposed methods via extensive simulation studies and a real application to a microarray data set.