Machine-guided path sampling to discover mechanisms of molecular self-organization

Area of Science:

• Computational chemistry
• Biophysics
• Materials science

Background:

• Molecular self-organization is fundamental to life and materials.
• Understanding the mechanisms of these processes is challenging due to complex many-body interactions.

Purpose of the Study:

• To present an autonomous path sampling algorithm for discovering molecular self-organization mechanisms.
• To integrate deep learning and transition path theory for enhanced sampling of rare events.

Main Methods:

• An autonomous algorithm using deep learning and transition path theory.
• A learning cycle that constructs, validates, and updates mechanistic models.
• Symbolic regression to distill mechanisms into interpretable physical observables.

Main Results:

• Successfully applied to ion association, gas-hydrate formation, polymer folding, and membrane-protein assembly.
• Captured many-body solvent motions governing assembly.
• Identified key variables for classical nucleation theory.
• Revealed competing assembly pathways and folding mechanisms at various resolutions.

Conclusions:

• The developed algorithm effectively elucidates complex molecular self-organization mechanisms.
• Mechanistic descriptions are transferable across thermodynamic states and chemical spaces.
• Provides a powerful tool for studying diverse self-assembling systems.