MarZIC: A Marginal Mediation Model for Zero-Inflated Compositional Mediators with Applications to Microbiome Data

Area of Science:

• Microbiome Research
• Causal Inference
• Statistical Genetics

Background:

• The human microbiome plays a role in complex disease development through mediating causal pathways.
• Standard mediation analysis struggles with microbiome data due to excessive zeros and compositional constraints.
• Key challenges include analyzing the mediation effect of zero-inflated data and identifying false zeros.

Purpose of the Study:

• To develop a novel statistical method for analyzing microbiome mediation effects.
• To address the challenges posed by zero-inflated and compositional microbiome data.
• To provide a robust framework for causal mediation analysis in microbiome studies.

Main Methods:

• Developed a marginal mediation analysis method within the potential-outcomes framework.
• The method accounts for the compositional nature of microbiome data.
• Utilized probabilistic models to address zero-inflation and false zeros.

Main Results:

• The proposed marginal model successfully decomposes mediation effects inherent to zero-inflated distributions.
• Probabilistic modeling effectively handles the issue of false zeros.
• Simulations and a real-world microbiome study demonstrated superior performance compared to existing methods.

Conclusions:

• The novel MarZIC approach demonstrates superior performance in analyzing zero-inflated microbiome composition as mediators.
• MarZIC outperforms standard causal mediation analysis and other competing methods.
• This method offers a more accurate way to understand microbiome-mediated disease pathways.