A0533
Title: High dimensional tests on multivariate regressions under confounding
Authors: Shota Katayama - Keio University (Japan) [presenting]
Abstract: In high dimensional data analysis, especially in differential gene expression analysis, detecting the difference of a huge number of features between two groups is an essential problem. A unified inference on high dimensional parameters is provided for comparing two groups in the case where random assignment is not feasible. To achieve this, multivariate regressions with high dimensional response vector is considered, taking into account observable confounding variables. Based on an efficient score function, global and multiple testing procedures on its high dimensional parameter are proposed to ensure asymptotic validity in high dimensions. Applying real RNA-seq data on Covid-19 patients demonstrates significant genes involved in serious cases.