MLHessian-TSopt

This repository contains wrapper scripts for running and analyzing transition state calculations using NewtonNet deep learning model and Sella optimizer.

Script overview

• Scripts/split.ipynb: Wrapper notebook for model training data splitting from Transition-1x dataset with augmentation from ANI-1x dataset.
• Scripts/test.ipynb: Wrapper notebook for model testing using the holdout test reactions in Transition-1x dataset.
• Scripts/noise.ipynb: Wrapper notebook for initial guess geometry generation and subsequent noising of Sella benchmark reactions.
• Scripts/opt/nn_sella_quacc.py: Wrapper script for NewtonNet-based optimizations.
• Scripts/opt/dft_sella_quacc.py: Wrapper script for DFT (Density Functional Theory) method, specifically using the wb97x/6-31G* level of theory.
• Scripts/gather.ipynb: Wrapper notebook for optimization data retrieval.

Models

• Models/PretrainedModels/training_1: Pre-trained model in NewtonNet paper, trained on ANI dataset.
• Models/PretrainedModels/training_9: Pre-trained model in NewtonNet paper, trained on ANI-1x dataset.
• Models/FinetunedModels/training_44: Fine-tuned model from training_1 above, trained on Transition-1x dataset composition split 5.
• Models/FinetunedModels/training_56: Fine-tuned model from training_1 above, trained on Transition-1x dataset composition split 5.
• Models/FinetunedModels/training_64: Fine-tuned model from training_1 above, trained on Transition-1x dataset composition split 51.
• Models/FinetunedModels/training_63: Fine-tuned model from training_1 above, trained on Transition-1x dataset composition split 52.
• Models/FinetunedModels/training_58: Fine-tuned model from training_1 above, trained on Transition-1x dataset composition split 53.
• Models/FinetunedModels/training_52: Fine-tuned model from training_1 above, trained on Transition-1x dataset conformation split 0.
• Models/FinetunedModels/training_54: Fine-tuned model from training_1 above, trained on Transition-1x dataset conformation split 1.
• Models/FinetunedModels/training_53: Fine-tuned model from training_1 above, trained on Transition-1x dataset conformation split 2.
• Models/FinetunedModels/training_55: Fine-tuned model from training_1 above, trained on Transition-1x dataset conformation split 3.

Analysis

• Analysis/Figure1b.ipynb: Wrapper notebook for dataset interatomic distance distribution analysis.
• Analysis/Figure1cd.ipynb and Analysis/FigureS1.ipynb: Wrapper notebook for model testing regarding energy and force predictions.
• Analysis/Figure2.ipynb and Analysis/Figure4bc.ipynb: Wrapper notebook for model testing regarding Hessian predictions.
• Analysis/Figure3.ipynb: Wrapper notebook for optimization path comparisons.
• Analysis/Figure4c.ipynb: Wrapper notebook for optimized transition state comparisons.
• Analysis/Figure4abd.ipynb: Wrapper notebook for optimized reactant/product comparisons.

Note

For detailed information on setup and configuration, please refer to the following:

1. Sella Package:

   • GitHub Repository
   • Documentation

2. NewtonNet:

   • GitHub Repository

3. QuAcc Recipes for NewtonNet and QChem:

   • GitHub Repository
   • Documentation

4. Corresponding Paper Authors:

   • Feel free to reach out to them (including me) for assistance.