Adaptive testing on a high-dimensional parameter with application to precision medicine

Thursday, November 8, 2018 - 2:00pm - 2:30pm
Lind 305
Wei Pan (University of Minnesota, Twin Cities)
A key issue in precision medicine is to uncover and utilize interactions between treatments and one's genetic and environmental risk factors. Due to the polygenic nature, testing for the interaction between a treatment and one's genetic features like a set of SNPs involves testing on a high-dimensional parameter. For such a purpose, it is critical to apply a test that can maintain high power across various scenarios. We demonstrate the issue under a simple normal model for one-sample testing: while the LRT/Wald/score tests are equivalent with diminishing power under some general conditions, our proposed adaptive sum of powered score (aSPU) test is usually much more powerful.