Unsupervised synchronization of molecular dynamics trajectories via graph embedding and time warping

Area of Science:

• Computational Biology
• Biophysics
• Structural Bioinformatics

Background:

• Molecular dynamics (MD) simulations offer atomistic detail but comparing independent trajectories is difficult due to divergence.
• Misaligned simulations hinder reproducibility and accurate mechanistic interpretation of biomolecular processes.
• A generalizable, unsupervised method is needed to synchronize and compare MD trajectories across diverse systems and conditions.

Purpose of the Study:

• To develop and present NetMD, a novel computational framework for synchronizing and analyzing molecular dynamics trajectories.
• To enable robust comparison of MD simulation ensembles across heterogeneous conditions.
• To facilitate the uncovering of conserved patterns and critical divergences in biomolecular dynamics.

Main Methods:

• NetMD integrates graph-based representations with dynamic time warping for trajectory analysis.
• Trajectory frames are converted into entropy-filtered residue-contact graphs and embedded as low-dimensional vectors.
• Time-warping barycenter averaging aligns vectorized trajectories, generating a consensus trajectory and pruning outliers.

Main Results:

• NetMD was applied to transporters, demethylases, and protein complexes relevant to neurological disease and cancer.
• The framework revealed shared multiphase dynamics and identified mutation- or ligand-specific deviations in biomolecular systems.
• NetMD demonstrated its ability to enable direct, time-resolved comparison of MD ensembles.

Conclusions:

• NetMD provides an unsupervised, broadly applicable tool for analyzing and comparing molecular dynamics simulations.
• The framework enhances reproducibility and mechanistic interpretation by synchronizing trajectories.
• NetMD facilitates the discovery of conserved biomolecular mechanisms and system-specific variations.