time-structured-vae

This repository includes deep-learning methods for obtaining slow collective variables (CVs) for biomolecules. The included methods are following:

Method	Reference
TAE	Wehmeyer, C.; Noé, F. Time-lagged autoencoders: Deep learning of slow collective variables for molecular kinetics. J. Chem. Phys. 2018, 148, 241703.
TVAE	Hoffmann, M.; Scherer, M. K.; Hempel, T.; Mardt, A.; de Silva, B.; Husic, B. E.; Klus, S.; Wu, H.; Kutz, J. N.; Brunton, S.; Noé, F. Deeptime: a Python library for machine learning dynamical models from time series data. Machine Learning: Science and Technology 2021.
VDE	Hernándeza, C. X.; Wayment-Steele, H. K.; Sultan, M. M.; Husic, B. E.; Pande, V. S. Variational encoding of complex dynamics. Phys. Rev. E 2018, 97, 062412.
SRV	Chen, W.; Sidky, H.; Ferguson, A. L. Nonlinear Discovery of Slow Molecular Modes using State-Free Reversible VAMPnets J. Chem. Phys. 2019, 150, 214114.
tsVAE (ours)	Ishizone, T.; Matsunaga, Y.; Fuchigami, S.; Nakamura, K. Representation of Protein Dynamics Disentangled by Time-Structure-Based Prior. J. Chem. Theory Comput. 2023, 20, 1, 436--450
tsTVAE (ours)	Ishizone, T.; Matsunaga, Y.; Fuchigami, S.; Nakamura, K. Representation of Protein Dynamics Disentangled by Time-Structure-Based Prior. J. Chem. Theory Comput. 2023, 20, 1, 436--450

Requirement

We use following environment:

• numpy 1.19.5
• msmbuilder 3.8.0
• torch 1.10.1
• mdshare 0.4.2 (optional for fetching data)
• pyemma 2.5.7 (optional for MSM analysis)
• matplotlib 3.3.4 (optional for graphics)
• scikit-learn 0.18 (optional for preprocessing)
• scipy 1.2.1 (used in SRV, tsVAE, and tsTVAE)

Usage

You can use deep-learning methods by

import models
m = models.tsTVAE(input_dim=30, lagtime=1, n_epochs=10, latent_dim=2)
embed = m.fit_transform(data)

The example of the application is shown in example.ipynb.

Citation

If you use this code in your work, please cite:

@article{tsvae,
  title = {Representation of Protein Dynamics Disentangled by Time-Structure-Based Prior},
  author = {Ishizone, Tsuyoshi and Matsunaga, Yasuhiro and Fuchigami, Sotaro and Nakamura, Kazuyuki},
  journal = {J. Chem. Theory Comput.},
  volume = {20},
  issue = {1},
  pages = {436--450},
  numpages = {15},
  year = {2024},
  month = {1},
  publisher = {American Chemical Society},
  url = {https://pubs.acs.org/doi/full/10.1021/acs.jctc.3c01025}
}