This repository deals with the task of predicting activation energy of reactions from fixed-width vector representations ("fingerprints") of the reactant and product molecules from their SMILES (Simplified Molecular Input Line-Entry System) representations. Simple regression models such as linear, ridge, LASSO and support vector regression are compared against more advanced models like random forests, gradient-boosted random forests and multi-layered perceptrons.

Two datasets are used in this study:

1. Grambow et al (2020) (https://www.nature.com/articles/s41597-020-0460-4)
2. Reaction Graph Depth (RGD) Dataset (https://engineering.purdue.edu/savoiegroup/data+code.html)

pytorch
pytorch-lightning
hydra
scikit-learn
xgboost
seaborn
matplotlib
numpy
rdkit
joblib

To run, first create the environment using either pip or conda (using requirements.txt or environment.yml, respectively, from the environment directory of this repo). Then, to run linear regression on the Grambow dataset with the default settings, run

python run_sklearn.py

To run regression with the PyTorch MLP using the default settings, run

python run_pytorch.py

This project uses Hydra (https://github.com/facebookresearch/hydra) for configuration management. To learn more on how to use Hydra, please refer to documentation in Hydra's GitHub page.