Prediction of Long-Time Molecular Motions Using MD-GAN

By using “MD-GAN” (developed by the Yasuoka Laboratory at Keio University), which integrates machine learning with simulation technology, it is possible to predict long-time molecular motions from short-time molecular simulation results. This enables efficient investigation of long-time changes such as molecular diffusion without performing time-consuming calculations.

Use Cases Highlights

Prediction of long-time molecular motion from short-time MD results using machine learning
Effective for evaluating molecular diffusivity
Utilization as a surrogate model for molecular motion

Prediction of long-time molecular motion and evaluation of diffusion phenomena from short-time MD results using machine learning

Results of predicting long-time molecular motion from short-time data using MD-GAN are shown. This enables accurate prediction of long-time phenomena such as diffusion from short MD simulations.

Prediction of long-time molecular motion by MD-GAN

Effective for evaluating molecular diffusivity

A graph of the mean square displacement (MSD) of water is shown. The black dashed line represents MD calculation results, the red solid line represents predictions from MD-GAN, and the green region indicates short-time data used for training. The prediction results in the white long-time region also match the MD calculations well, demonstrating the high predictive accuracy of MD-GAN.

MSD predicted by MD-GAN

Reference

[1] K. Endo, K. Tomobe, and K. Yasuoka, In Thirty-Second AAAI Conference on Artificial Intelligence, pp. 2192-2199, (2018)
https://ojs.aaai.org/index.php/AAAI/article/view/11863/11722

Details of analysis

Related information