mbDenoise: microbiome data denoising using zero-inflated probabilistic principal components analysis.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Microbiome data analysis faces significant technical hurdles, including a high proportion of zero counts (both biological and technical zeros).
• Measurements are further complicated by unequal sequencing depth, overdispersion, and data redundancy, introducing substantial noise that obscures true biological signals.

Purpose of the Study:

• To introduce mbDenoise, an accurate and robust method for denoising microbiome count data.
• To address the challenges of zero-inflation, unequal sequencing depth, and noise in microbiome datasets.

Main Methods:

• Developed mbDenoise, a novel method based on a zero-inflated probabilistic principal component analysis (ZIPPCA) model.
• Employed variational approximation to learn the latent data structure and recover true abundance levels via the posterior distribution.
• Leveraged information sharing across samples and taxa for improved accuracy.

Main Results:

• mbDenoise effectively denoises microbiome data, reducing noise introduced by technical factors.
• The method successfully recovers true abundance levels by learning the underlying latent structure.
• mbDenoise demonstrates superior performance compared to existing state-of-the-art methods in signal extraction.

Conclusions:

• mbDenoise provides an accurate and robust solution for denoising microbiome data.
• The method enhances the reliability of microbiome data for downstream analyses by mitigating technical noise.
• mbDenoise represents a significant advancement in the computational analysis of microbiome datasets.