MarkerML

MarkerML identifies environment-specific marker features in metagenomic and microbiome datasets to distinguish contrasting microbial community states.

Key Features:

• Interpretable Machine Learning: Employs interpretable machine learning with Shapley Additive Explanations (SHAP) to quantify feature importance and inter-dependence.
• Automated Analysis: Automates the application of SHAP in conjunction with compositionality-aware hypothesis testing for multivariate microbiome data.
• Visualization Outputs: Produces prediction effect plots, model performance reports, feature dependency plots, Shapley-informed cladograms (Sungrams), and hypothesis-tested violin plots.
• Bias Mitigation and Reproducibility: Includes provisions for excluding participant bias and ensuring reproducibility of results.
• Broad Applicability: Focuses on microbiome studies while retaining potential applicability to other biological domains.

Scientific Applications:

• Marker Feature Identification: Identification of environment-specific marker features that distinguish contrasting or comparable microbial community states in metagenomic datasets.
• Microbial Ecology and Health Studies: Supports hypothesis-driven investigation of microbial ecology, health, disease, and environmental impacts by providing interpretable feature signals.
• Multivariate Microbiome Analysis: Enables analysis of multivariate microbiome datasets to elucidate the roles and dependencies of marker features.

Methodology:

Applies interpretable machine learning using Shapley Additive Explanations (SHAP) together with compositionality-aware hypothesis testing, automates SHAP application, and includes provisions to exclude participant bias and ensure reproducibility.