A1033
Title: Poisson principal component analysis and its ensemble approaches for cross-study analyses
Authors: Hong Gu - Dalhousie University (Canada) [presenting]
Toby Kenney - Dalhousie University (Canada)
Molly Hayes - Dalhousie University (Canada)
Tianshu Huang - Dalhousie University (Canada)
Abstract: High-dimensional count data are ubiquitous. Parametric methods based on log-normal Poisson distribution assumptions for principal component analysis (PCA) are typically sensitive to outliers. The aim is to first present a semi-parametric PCA (Poisson PCA) method generally applicable to count data for dimension reduction and data exploration, then further present a family of Poisson PCA ensemble methods for common principal component or common factor analysis approaches to cross-study analyses on multiple data sets. Applications using microbiome count data to find microbial communities are used to demonstrate the methods.
