mbDenoise

mbDenoise applies a zero-inflated probabilistic principal components analysis (ZIPPCA) framework to denoise microbiome count matrices by distinguishing biological zeros from technical artifacts and recovering true taxa abundances for accurate downstream analyses.

Key Features:

• Zero-Inflated Probabilistic PCA (ZIPPCA) Model: Uses a zero-inflated probabilistic PCA model to separate biological zeros from technical zeros in microbiome count data.
• Variational Approximation: Fits the latent-structure model via variational approximation to approximate posterior distributions efficiently.
• Posterior-Based Recovery: Recovers true abundance levels across samples and taxa by leveraging the model posterior distribution.
• Borrowing Information Across Samples and Taxa: Shares information across samples and taxa during posterior inference to improve reliability of abundance estimates.
• Superior Performance: Demonstrates improved ability to extract meaningful signals from noisy microbiome datasets compared to existing methods.

Scientific Applications:

• Microbial Community Analysis: Produces denoised abundance estimates to clarify community composition and dynamics.
• Health and Disease Associations: Provides more accurate taxa abundance inputs for studying associations between microbiome profiles and health conditions.
• Ecological Studies: Reduces noise that can obscure ecological interactions within microbial ecosystems, supporting ecological inference.

Methodology:

Fits a zero-inflated probabilistic PCA (ZIPPCA) model using variational approximation and performs posterior-based recovery of abundances by borrowing information across samples and taxa.