MDGraphEmb: a toolkit for graph embedding and classification of protein conformational ensembles.

Area of Science:

• Computational Biology
• Biophysics
• Machine Learning

Background:

• Molecular Dynamics (MD) simulations are crucial for studying protein dynamics and function.
• Integrating MD simulations with machine learning faces challenges in optimal data representation.
• Graph embedding offers a method to learn low-dimensional representations from graph structures, bridging simulation data and machine learning.

Purpose of the Study:

• To introduce MDGraphEmb, a Python library for converting protein MD simulation trajectories into graph representations and embeddings.
• To enable compression of high-dimensional, noisy MD data into machine-learning-ready tabular formats.
• To facilitate the analysis of protein dynamics and identification of important protein conformations.

Main Methods:

• Developed MDGraphEmb, a Python library built upon MDAnalysis.
• Implemented conversion of protein MD trajectories into graph-based representations.
• Supported various graph embedding techniques and machine learning models for workflow creation.
• Applied graph embedding to encode and detect protein conformations for functional state classification.

Main Results:

• MDGraphEmb effectively captures and compresses structural information from MD simulation data.
• Graph embeddings are suitable for diverse downstream machine learning classification tasks.
• Demonstrated utility in classifying functional states of adenylate kinase (ADK), Plantaricin E (PlnE), and HIV-1 protease.
• Provided a performance comparison of different graph embedding methods and machine learning models.

Conclusions:

• Graph embedding provides an efficient method for representing and analyzing protein dynamics from MD simulations.
• MDGraphEmb facilitates the application of machine learning to understand protein conformational changes and functional states.
• The library's framework supports the development of novel workflows for computational biophysics research.