ML-VEG

This repository contains the data and the scripts used for the manuscript "Machine Learning Approach To Vertical Energy Gap in Redox Processes".

Building conda environment

We suggest running the following command to create a conda environment called VEGPred:

conda env create -f environment.yml

Repo details

Features:

• The feature list calculated using Hartree-Fock (HF) and semi-empirical (EMP1) methods for each system are in their respective folders. For example,
  1. The HF features for QM cutoff 0.0 Å for benzene are in the benzene/feature_list_0.0.dat folder.
  2. The EMP1 features for QM cutoff 7.5 Å for lumiflavin are in the lumiflavin/EMP1_feature_list_7.5.dat folder.

Workflow:

• The workflow is organized as follows:
  1. Add ML models with parameter list in the models_hyperparam_opt.py file.
  2. Run models_hyperparam_opt.py > model_list.dat to create list of best models for each system.
  3. Select one ML model from model_list.dat and add it to the system dictionary in opt_model_test.py .
  4. Run save_models_and_plot.py to produce heatmaps, parity plots, and learning curves as presented in the manuscript.