Training and simulation inputs for paper "Descriptors-free Collective Variables From Geometric Graph Neural Networks".
arXiv link to the paper: https://arxiv.org/abs/2409.07339v1
The contents are organized as follows:
- plumed_pytorch_gnn: contains the plumed interface for GNN-based CVs
- alad: contains the files to reproduce alanine dipeptide in a vacuum results
- data: topology and force field files
- analysis: analysis scripts
- train: scripts for the training of the GNN-CVs and trained model
- run_biased: simulation files for biased simulations using phi and psi as CVs
- run_biased_gnn/10A_2layer_1c/1: simulation files for biased simulations using GNN-CV
- run_unbiased/long: simulation files for unbiased simulations
- nacl: contains the files reproduce NaCl dissociation in explicit water results
- data: topology and force field files
- analysis: analysis scripts
- train: scripts for the training of the GNN-CVs and trained model
- run_biased: simulation files for biased simulations using interionic distance and oxygen coordination of Na+ as CVs
- run_biased_gnn/6A_2layer_1c/1: simulation files for biased simulations using GNN-CV
- run_unbiased/long: simulation files for unbiased simulations
- reaction: contains the files reproduce methyl migration of FDMB cation results
- data: topology files
- eval: plumed files to evaluate GNN-CV using plumed driver
- run_biased: simulation files for biased simulations using coordiantion difference as CV
- run_biased_gnn: simulation files for biased simulations using GNN-CV
- run_unbiased: simulation files for unbiased simulations
- train_ff: scripts for the training of MLCV based on feed-forward NN and trained model
- train_ff_full: scripts for the training of MLCV based on feed-forward NN and a fully permutated dataset and trained model
- train_gnn: scripts for the training of the GNN-CVs and trained model
The modified version of the mlcolvar library used for the GNN-CV training is available at: https://github.com/jintuzhang/mlcolvar
The relevant code for the GNN-CV definiton and training is implemented in the mlcolvar.graph module of such a library.