mdxplain: Scalable molecular dynamics analysis with machine learning-based feature selection and modular workflows

Area of Science:

• Computational chemistry
• Biophysics
• Data science

Background:

• Molecular dynamics (MD) simulations offer detailed insights into molecular motion.
• Increasingly large datasets from MD simulations necessitate efficient analysis tools.

Purpose of the Study:

• To introduce mdxplain, a Python API for scalable analysis of large MD datasets.
• To facilitate the identification of structural patterns and dynamic behaviors in molecular systems.

Main Methods:

• Developed a high-level Python API (mdxplain) with a unified object for MD featurization, metrics, dimensionality reduction, clustering, and feature selection.
• Implemented optimized memory handling for large datasets (millions of frames) and support for multiple topologies.
• Integrated decision trees for feature selection and PyMOL/NGLView for 3D visualization.

Main Results:

• mdxplain streamlines the creation of reusable analysis pipelines with minimal code.
• The tool efficiently processes large MD datasets, identifying key structural and dynamic patterns.
• Generated reports include plots, representative conformations, decision trees, and 3D visualizations.

Conclusions:

• mdxplain empowers both expert and non-expert users to analyze complex MD data effectively.
• The API ensures reproducibility and FAIR compliance through exportable pipelines and bundled data.
• mdxplain is available on GitHub with comprehensive documentation and tutorials.