A0938
Title: Zero-inflated factor analysis skew normal for microbiome dimension reduction
Authors: Kevin McGregor - University of Manitoba (Canada) [presenting]
Abstract: Advancements in next-generation sequencing have transformed the understanding of host-microbe interactions, revealing links between microbial composition and chronic conditions such as obesity, diabetes, IBD, etc. However, analyzing microbiome data presents challenges due to high dimensionality, sparsity, overdispersion, and its compositional nature. Existing models for dimension reduction, such as ZIPPCA-LNM and ZIPPCA-LPNM, offer a flexible probabilistic framework for composition estimation. However, capturing the high skewness inherent in log-ratio transformed microbiome data remains a challenge. To address these issues, a zero-inflated factor analysis skew-normal (ZIFA-SN) model is proposed, incorporating a skew-normal distribution as a prior for latent factors, effectively capturing data asymmetry. A mean-field variational Bayes approach is employed to approximate the intractable posterior, offering a computationally efficient alternative to traditional MCMC methods.
