Related Experiment Video
Updated: May 21, 2025
10:58
Protein WISDOM: A Workbench for In silico De novo Design of BioMolecules
Published on: July 25, 2013
17.0K
Unsupervised Learning of Progress Coordinates during Weighted Ensemble Simulations: Application to NTL9 Protein
Jeremy M G Leung1, Nicolas C Frazee1, Alexander Brace2,3
1Department of Chemistry, University of Pittsburgh, Pittsburgh, Pennsylvania 15260, United States.
Journal of Chemical Theory and Computation
|March 19, 2025
Summary
We developed a deep learning (DL) method to identify key molecular motions for rare-event sampling. This approach enhances weighted ensemble (WE) simulations by over threefold, accelerating the study of protein folding.
Area of Science:
- Computational Chemistry
- Biophysics
- Machine Learning
Background:
- Identifying slow molecular motions is crucial for rare-event sampling strategies.
- Unsupervised machine learning methods are sought after for discovering these progress coordinates.
Purpose of the Study:
- To develop a general, unsupervised method for identifying progress coordinates on-the-fly during weighted ensemble (WE) simulations.
- To enhance the efficiency of rare-event sampling using deep learning (DL).
Main Methods:
- Implemented a DL approach using a convolutional variational autoencoder to identify outliers in sampled conformations.
- Integrated DL outlier identification into the WE sampling workflow.
- Utilized synthetic molecular dynamics trajectories from a Markov state model for efficient testing.
Main Results:
- The DL-enhanced WE method significantly improved sampling efficiency by over threefold for estimating the NTL9 protein folding rate constant.
- Demonstrated the method's capability in unsupervised learning of slow coordinates.
Conclusions:
- This work presents a significant advancement in unsupervised learning for rare-event sampling.
- The DL-enhanced WE method offers a powerful tool for molecular simulations, particularly for complex processes like protein folding.
