The mixed model for the analysis of a repeated-measurement multivariate count data

Area of Science:

• Statistical modeling
• Microbiome analysis
• Bioinformatics

Background:

• Clustered, overdispersed multivariate count data present modeling challenges due to intra- and inter-sample correlations.
• Quantifying correlation between time points and assessing covariate effects on count distributions are crucial.
• Microbiome data, particularly 16S rRNA gene amplicon sequencing data, exhibit compositional structure and overdispersion.

Purpose of the Study:

• To extend the Dirichlet-multinomial distribution regression model using random effects to handle clustering in multivariate count data.
• To develop a Dirichlet-multinomial mixed regression model for analyzing microbiome data.
• To compare this approach with a negative binomial regression mixed model.

Main Methods:

• Development and application of the Dirichlet-multinomial mixed regression model.
• Incorporation of subject-specific and categorical-specific random effects.
• Analysis of microbiome data from an Indonesian epidemiological study.

Main Results:

• Time was found to have no statistically significant effect on microbiome composition.
• Correlation between subjects was statistically significant, indicating inter-individual variability.
• Treatment showed a significant effect on microbiome composition exclusively in infected subjects who remained infected.

Conclusions:

• The Dirichlet-multinomial mixed regression model effectively addresses complex correlations and overdispersion in microbiome data.
• Subject-specific correlations are significant in microbiome composition.
• Treatment interventions impact microbiome composition differently based on infection persistence.