STROBE-causal machine learning for the human microbiome: systematic review on methodological innovations and...

Area of Science:

• Microbiome research
• Causal inference
• Machine learning

Background:

• The reproducibility crisis in causal microbiome research stems from inconsistent validation methods, poor interpretability, and lack of standardized reporting.
• This hinders reliable causal inference in microbiome-host interaction studies.

Purpose of the Study:

• To establish benchmarking standards for causal microbiome research using synthetic data and biological plausibility.
• To compare advanced causal machine learning (ML) methods like Double/Debiased ML, Deep IV, and DAGs for microbiome-host systems.
• To propose STROBE-CML guidelines for standardizing reporting in causal ML for microbiome research.

Main Methods:

• Systematic review of over 60 peer-reviewed studies (2015-2024).
• Evaluation of causal ML methodologies including Double/Debiased ML, Deep IV, and DAGs.
• Development of benchmarking standards and STROBE-CML reporting guidelines.

Main Results:

• Identified gaps in validation, interpretability, and reporting in current causal microbiome research.
• Highlighted innovations like federated validation and time-series causal discovery for reproducible inference.
• Introduced a decision support tool for selecting appropriate causal ML approaches.

Conclusions:

• The proposed benchmarking standards, STROBE-CML guidelines, and decision support tool offer a roadmap for reliable causal inference in microbiome science.
• These advancements facilitate biologically interpretable and clinically translatable causal claims.
• Addressing the reproducibility crisis requires robust validation and standardized reporting in causal microbiome research.