Presented By: Department of Statistics
Statistics Department Seminar Series: Michael Newton, Professor, Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison
On clustering to improve power in multiple hypothesis testing
Abstract:
Contemporary applications of statistics continue to fuel research in methodologies for high-dimensional hypothesis testing. One approach to increase the amount of data (and thus power) for a unit on test is to merge data from other units having similar data characteristics. I will present one version of this approach in the context where units are associated with nodes of an undirected graph; I will present findings on the sampling properties of the test statistics, connections to Bayesian analysis, and preliminary numerical results. I will also discuss a related clustering/testing problem from the analysis of single-cell RNA-Seq data and a model-based solution in this case.
Contemporary applications of statistics continue to fuel research in methodologies for high-dimensional hypothesis testing. One approach to increase the amount of data (and thus power) for a unit on test is to merge data from other units having similar data characteristics. I will present one version of this approach in the context where units are associated with nodes of an undirected graph; I will present findings on the sampling properties of the test statistics, connections to Bayesian analysis, and preliminary numerical results. I will also discuss a related clustering/testing problem from the analysis of single-cell RNA-Seq data and a model-based solution in this case.
Explore Similar Events
-
Loading Similar Events...
