A1001
Title: Bayesian dimension reduction in microbiome platforms
Authors: Kevin McGregor - University of Manitoba (Canada) [presenting]
Abstract: Dimension reduction techniques are among the most essential analytical tools in analyzing high-dimensional data. Generalized principal component analysis is an extension to standard principal component analysis (PCA) for various types of non-Gaussian data and has been widely used to identify low-dimensional features in high-dimensional data. For microbiome count data, the multinomial PCA is a natural counterpart to standard PCA. However, this technique fails to account for the excessive number of zero values frequently observed in microbiome count data. To allow for sparsity, zero-inflated multivariate distributions can be used. A Bayesian zero-inflated probabilistic PCA model is proposed for extracting information in compositional count data. A classification variational approximation algorithm is developed to fit the model. A simulation study and an application in a pediatric-onset multiple sclerosis metagenomic dataset will be further featured.
