Zero-inflated Poisson factor model with application to microbiome read counts.

Area of Science:

• Microbiome Bioinformatics
• Statistical Modeling
• High-Dimensional Data Analysis

Background:

• High-dimensional microbiome data presents unique challenges for analysis, including count-valued nature and excessive zero reads.
• Existing dimension reduction techniques often fail to adequately address these specific data characteristics.
• Accurate dimension reduction is crucial for subsequent analyses like regression and clustering in microbiome studies.

Purpose of the Study:

• To propose a novel statistical model for dimension reduction of high-dimensional microbiome data.
• To specifically address the challenges posed by count-valued data and excessive zero reads.
• To provide interpretable low-dimensional representations for further microbiome data analysis.

Main Methods:

• Development of a zero-inflated Poisson factor analysis model.
• The model incorporates library size as an offset and relates Poisson rates to zero occurrences.
• An efficient expectation-maximization algorithm was developed for robust parameter estimation.

Main Results:

• The proposed zero-inflated Poisson factor analysis model effectively reduces dimensionality while accommodating data features.
• Simulation studies demonstrated the efficacy and robustness of the developed method.
• Application to a real-world study revealed insights into subgingival microbiome and periodontal disease.

Conclusions:

• The proposed zero-inflated Poisson factor analysis model offers a powerful tool for high-dimensional microbiome data.
• This method enhances the interpretability of microbiome data through low-dimensional scores and loadings.
• The approach facilitates a deeper understanding of the relationship between microbiome composition and disease states.