A1573
Title: Generalized functional probabilistic principal component analysis for longitudinal microbiome data analysis
Authors: Xiangnan Xu - Humboldt University of Berlin (Germany) [presenting]
Abstract: Longitudinal microbiome studies are essential for understanding the dynamic microbial communities that inhabit various body sites and their interactions with host health, offering valuable insights for precision medicine. However, the analysis of longitudinal microbiome data presents challenges due to its high dimensionality, compositionality, overdispersion, and typically sparse sampling at irregular time points. Existing methods for dimension reduction, such as functional data analysis and tensor decomposition, often assume uniform sampling across individuals or focus on Gaussian-distributed variables. To address these, a novel generalized functional probabilistic principal component analysis (GFPPCA) framework is proposed that extends functional tensor decomposition to the setting of the exponential family distribution. GFPPCA integrates functional tensor decomposition with generalized probabilistic principal component analysis, enabling efficient and interpretable dimension reduction for complex functional microbiome data. An efficient alternating direction method of the multipliers-based algorithm is developed to estimate model parameters and validate GFPPCA on both simulated and real-world infant microbiome datasets. Results demonstrate that GFPPCA robustly and efficiently captures the primary information of the data across various parameter settings.
