Skip to content

jjgoings/chem_interp

Repository files navigation

chem_interp: use interpretable machine learning to explain complicated chemical reactions!

Usually molecular dynamics has so much information, it's tough to make sense of it all. But what if you could have a computer help you figure out what's really going on? That's the idea here -- have the computer learn what molecular motions lead to chemical reactions, and then distill this knowledge in a way humans can understand. First the computer learns, then it teaches.

Code was used in this paper:

Goings, J. J.; Hammes-Schiffer, S. Nonequilibrium Dynamics of Proton-Coupled Electron Transfer in Proton Wires: Concerted but Asynchronous Mechanisms. ACS Cent Sci 2020, 6 (9), 1594–1601.

Big-picture overview:

  1. Fit neural network to predict proton transfer times from ab initio molecular dynamics.
  2. Use interpretable ML methods to explain what molecular motions led to proton transfer.
Predict Explain

Top level has code for training the neural network. Once the model is trained, there are two types of approaches for explaining the model in explanations: permutation importance and SHAP (SHapley Additive exPlanations).

Permutation importance is subtractive -- it tells you how much your model breaks when you scramble the data for that particular feature.

SHAP is additive -- it tells you how much a feature contributed to a model prediction.

If the two approaches identify features as being important, there's probably good reason to investigate further.

Directory structure:

.
├── README.md
├── 1-prepare_model.py
├── 2-keep_training.py
├── 3-evaluate_model.py
├── data
│   ├── 1-split_data.py
│   ├── 2-scale_data.py
│   ├── processed
│   ├── raw
│   └── split_raw
├── figures
├── model
├── explanations
│   ├── permutation_importance
│   │   ├── 1-run_permutation_importance.py
│   │   ├── 2-plot_permutation_importance.py
│   │   ├── data
│   │   └── figures
│   └── shap
│       ├── 1-run_shap_explainer.py
│       ├── 2-plot_mode_impact_scatter.py
│       ├── 3-plot_mode_impact_magnitude.py
│       ├── 4-plot_impact_vs_displacement.py
│       ├── data
│       └── figures
└── utils.py

About

use interpretable machine learning to explain chemical reactions

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages