Biostatistics Seminar Series - Siyuan Ma, PhD

Title:  A Statistical Model for Microbial Abundances, with Applications in Simulation and TestingAbstract:  Many statistical methods have been developed for microbial community profiles, but due to the data's complex nature (zero-inflation, non-independence, and compositionality), it is difficult to evaluate such methods within a single framework. We developed SparseDOSSA: a statistical model for the joint distribution of microbial abundances. SparseDOSSA captures both marginal microbial feature abundances with zero-inflated log-normal distributions, as well as feature-feature interactions with a Gaussian copula. We demonstrate SparseDOSSA's application for 1) accurately modeling microbial population profiles, and 2) spiking-in true positive synthetic associations to benchmark analysis methods. In a follow-up project, we combine SparseDOSSA with the conditional randomization testing (CRT) framework, for FDR-controlled testing of microbial associations.