Machine learning from quantum chemistry to predict experimental solvent effects on reaction rates

Area of Science:

• Computational chemistry
• Chemical kinetics
• Machine learning applications

Background:

• Accurate prediction of solvent effects on reaction rates is vital for chemical process design and high-throughput screening.
• Existing machine learning models are limited by data scarcity, hindering generalizability across diverse reactions and solvents.

Purpose of the Study:

• To develop a generalizable machine learning model for predicting kinetic solvent effects.
• To predict solvation free energy and solvation enthalpy of activation for solution-phase reactions.

Main Methods:

• Generated a large dataset of over 28,000 neutral reactions and 295 solvents using the COSMO-RS method.
• Trained a machine learning model to predict solvation free energy (ΔΔG‡solv) and solvation enthalpy of activation (ΔΔH‡solv).
• Validated the model on unseen reactions and experimental data.

Main Results:

• Achieved mean absolute errors of 0.71 kcal mol⁻¹ for ΔΔG‡solv and 1.03 kcal mol⁻¹ for ΔΔH‡solv relative to COSMO-RS calculations.
• Provided reliable predictions of relative rate constants within a factor of 4 on experimental data.
• The model offers near-instantaneous predictions based on atom-mapped reaction and solvent SMILES strings.

Conclusions:

• The developed machine learning model effectively predicts kinetic solvent effects for a wide range of reactions and solvents.
• This tool can accelerate chemical process design and solvent screening by providing rapid and accurate predictions.
• The model overcomes data limitations, enabling broader applicability in computational chemistry and reaction kinetics.