Clustering microbiome data using mixtures of logistic normal multinomial models.

Area of Science:

• Bioinformatics
• Microbiome Research
• Computational Biology

Background:

• Microbiome sequencing generates high-dimensional, over-dispersed, and compositional count data.
• Analyzing compositional data is challenging due to its simplex nature.
• Existing logistic normal multinomial models offer flexibility but incur high computational costs via Bayesian inference.

Purpose of the Study:

• To develop a novel mixture of logistic normal multinomial models for microbiome data clustering.
• To improve the computational efficiency of parameter estimation for these models.

Main Methods:

• Development of a mixture of logistic normal multinomial models.
• Implementation of variational Gaussian approximations (VGA) for parameter estimation.
• Application of additive log-ratio transformation to map compositional data to Euclidean space.

Main Results:

• The proposed method effectively clusters microbiome data.
• Variational Gaussian approximations substantially reduce computational overhead.
• The method demonstrates performance on both simulated and real microbiome datasets.

Conclusions:

• The novel mixture model with VGA provides an efficient approach for microbiome data analysis.
• This method addresses the computational challenges of existing models.
• It offers a flexible and scalable solution for clustering microbiome taxa.