mAML

mAML performs automated machine learning to generate optimized and interpretable models for microbial-feature-based disease prediction and classification.

Key Features:

• Automated Model Building: Automates creation of machine learning models from microbial features for classification tasks.
• Optimization and Interpretability: Produces models optimized for performance while enabling interpretability of microbial feature contributions.
• Reproducibility: Delivers reproducible results across analyses and benchmarking experiments.
• Performance Evaluation: Validated across 13 benchmark datasets spanning binary and multi-class classification tasks.
• GMrepo ML Repository Integration: Integrated with the GMrepo ML repository containing 120 microbial classification tasks linked to 85 human-disease phenotypes derived from the GMrepo database.
• Support for Metagenomic and Amplicon Data: Uses data derived from 12,429 metagenomic samples and 38,643 amplicon samples as represented in the GMrepo ML repository.

Scientific Applications:

• Disease Prediction and Classification: Enables microbial-feature-based prediction and classification of disease phenotypes.
• Microbiome Research: Supports analysis of complex microbial data associated with human-disease phenotypes.
• Algorithm Development and Benchmarking: Facilitates development and benchmarking of machine learning algorithms using curated benchmark datasets and the GMrepo ML repository.

Methodology:

Processes two input files through pre-training operations followed by model training; the pipeline architecture and pre-training operations are depicted in a flowchart.