TB-ML

TB-ML predicts antimicrobial resistance in Mycobacterium tuberculosis from genomic data using machine learning models.

Key Features:

• Docker containerization: Encapsulates published and unpublished machine learning models and their dependencies within Docker containers to ensure consistent execution across computing environments.
• Encapsulated ML models: Packages diverse ML models for antimicrobial resistance prediction specific to Mycobacterium tuberculosis.
• Minimal container I/O standard: Implements a minimal container input/output specification to maximize interoperability between pre-processing and prediction components.
• Pre-processing pipelines: Provides Docker containers for pre-processing steps that convert raw genomic data into formats compatible with ML models.
• Container integration into workflows: Integrates pre-processing and prediction containers into complete analysis workflows for end-to-end genomic-to-prediction processing.
• Extensibility and benchmarking: Supports addition of new models and pipelines and enables ongoing benchmark comparisons of ML approaches.
• Reproducibility and consistency: Uses containerization to reduce variability and promote reproducible ML-based resistance predictions across systems.

Scientific Applications:

• Drug resistance prediction: Predicts antimicrobial resistance phenotypes in Mycobacterium tuberculosis from genomic sequence data using machine learning.
• Genomic data integration: Integrates diverse ML models and pre-processing pipelines to enable comprehensive analyses of M. tuberculosis genomic datasets.
• Benchmarking ML approaches for AMR: Facilitates comparison and evaluation of machine learning methods for antimicrobial resistance prediction.
• Treatment strategy support: Provides genomic-based resistance predictions that can inform treatment strategy considerations for tuberculosis.

Methodology:

Machine learning models and pre-processing workflows are encapsulated in Docker containers; pre-processing containers convert genomic data into model-compatible formats; components adhere to a minimal container I/O standard and are integrated into analysis workflows.