Framework for Learning Effective Dynamics for Molecular Systems (LED-Molecular)

Basic code implementation of the paper: Accelerated Simulations of Molecular Systems through Learning of their Effective Dynamics, PR. Vlachas, J. Zavadlav, M. Praprotnik, P. Koumoutsakos, J. Chem. Theory Comput. 2022, 18, 1, 538–549 https://doi.org/10.1021/acs.jctc.1c00809

The LED-Molecular employs the following neural architectures:

• (AEs) Autoencoders to capture the effective degrees of freedom
• (MDNs) Mixture density networks for stochastic dynamics and probabilistic decoding
• (RNNs) Recurrent neural networks to capture nonlinear markovian dynamics on the latent space (LSTMs and GRUs, etc.)

Requirements

Code requirements are provided in the requirements.txt file. The code has been compiled with python version 3.7.

Relevant Publications

[1] Accelerated Simulations of Molecular Systems through Learning of Effective Dynamics, PR. Vlachas, J. Zavadlav, M. Praprotnik, P. Koumoutsakos Journal of Chemical Theory and Computation 18 (1), 538-549, 2022

[2] Multiscale Simulations of Complex Systems by Learning their Effective Dynamics, PR. Vlachas, G. Arampatzis, C. Uhler, P. Koumoutsakos Nature Machine Intelligence, 2022.

[3] Backpropagation Algorithms and Reservoir Computing in Recurrent Neural Networks for the Forecasting of Complex Spatiotemporal Dynamics, Pantelis R. Vlachas, Jaideep Pathak, Brian R. Hunt, Themistoklis P. Sapsis, Michelle Girvan, Edward Ott, Petros Koumoutsakos Journal of Neural Networks, 2020.

[4] Data-driven forecasting of high-dimensional chaotic systems with long short-term memory networks, Pantelis R. Vlachas, Wonmin Byeon, Zhong Y. Wan, Themistoklis P. Sapsis and Petros Koumoutsakos Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 474 (2213), 2018.