Data sets and machine learning models for: Machine learning from quantum chemistry to predict experimental solvent effects on reaction rates

The datasets and final machine learning model files for the manuscript "Machine learning from quantum chemistry to predict experimental solvent effects on reaction rates". Citation should refer directly to the manuscript:

Chung, Y.; Green, W. H. Machine learning from quantum chemistry to predict experimental solvent effects on reaction rates. ChemRxiv 2023, doi: 10.26434/chemrxiv-2023-f20bg

To use the machine learning models, please refer to the sample files and instructions on https://github.com/yunsiechung/chemprop/tree/RxnSolvKSE_ML. 

Detailed information can be found in README.md file.

Details on the files

In the pretraining and finetuning set csv files, each column represents:

rxn_smiles: atom-mapped reaction SMILES

solvent_smiles: solvent SMILES

ddGsolv: solvation free energy of activation of a reaction-solvent pair at 298K in kcal/mol (main prediction target)

ddHsolv: solvation enthalpy of activation of a reaction-solvent pair at 298K in kcal/mol (main prediction target)

dGsolv_reactant: solvation free energy of reactant(s) at 298K in kcal/mol (additional feature)

dGsolv_product: solvation free energy of product(s) at 298K in kcal/mol (additional feature)

dHsolv_reactant: solvation enthalpy of reactant(s) at 298K in kcal/mol (additional feature)

dHsolv_product: solvation enthalpy of product(s) at 298K in kcal/mol (additional feature)

Data sets under 'RxnSolvKSE_dataset_v1.0.zip'

pretraining_set: contains the dataset used for pre-training

all_data: contains all calculated data

pretraining_rxn_solvent_ddGsolv_ddHsolv_with_features_all.csv: contains both main prediction targets and additional feature for reaction-solvent pairs

pretraining_solvent_info.csv: list of all solvents

pretraining_unique_rxn.csv: list of all reactions, both forward and reverse directions

chosen_500k_data: contains the chosen 500k data

pretraining_rxn_solvent_ddGsolv_ddHsolv_500k.csv: contains main prediction targets for reaction-solvent pairs

pretraining_features_react_prod_dGsolv_dHsolv_500k.csv: contains additional features for reaction-solvent pairs

finetuning_set: contains the dataset used for fine-tuning

all_data: contains all calculated data

finetuning_rxn_solvent_ddGsolv_ddHsolv_with_features_all.csv: constains both main prediction targets and additional features for reaction-solvent pairs. The rxn_key column indicates whether the reaction is bimolecular hydrogen abstraction (bihabs), unimolecular hydrogen migration (intrahabs), or radical addition to a multiple bond (raddition). The 'fwd' and 'rev' each indicate forward and reverse reactions.

finetuning_solvent_info.csv: list of all solvents

finetuning_unique_rxn.csv: list of all reactions, both forward and reverse directions

chosen_data: contains chosen data

finetuning_rxn_solvent_ddGsolv_ddHsolv_chosen.csv: contains main prediction targets for reaction-solvent pairs

finetuning_features_react_prod_dGsolv_dHsolv_chosen.csv: contains additional features for reaction-solvent pairs

experimental_set/expt_rxn_atom_mapped_smiles.csv: contains the atom-mapped reaction SMILES used for the experimental data.The original experimental data can be found at https://zenodo.org/record/7747557.

Machine learning model files under 'RxnSolvKSE_ML_model_files.zip'

Contains the Chemprop machine learning model files for predicting ddGsolv and ddHsolv for a reaction-solvent pair. It takes atom-mapped reaction SMILES and solvent SMILES as inputs.

To use these ML models, please refer to the sample files and instructions on https://github.com/yunsiechung/chemprop/tree/RxnSolvKSE_ML