B2037
Title: Tensor decomposition of longitudinal microbiome data
Authors: Siyuan Ma - Vanderbilt University Medical Center (United States) [presenting]
Hongzhe Li - University of Pennsylvania (United States)
Abstract: A few methods available for unsupervised dimension reduction of longitudinal microbial abundance observations are discussed. Existing ones do not fully observe the distribution characteristics of such data types, namely, zero-inflation, compositionality, and overdispersion. We present a tensor decomposition model for dimension reduction of longitudinal microbiome data, by generalizing existing approaches in Gaussian data. Optimization is performed through projected gradient descent, additionally allowing interpretability constraints. Simulation studies show our method can recover low-rank structures in microbiome time course better than existing approaches. We applied our method to two existing longitudinal microbiome studies, to detect global microbial changes associated with dietary and pharmaceutical effects, as well as infant birth modes.
