Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2024 Jan 10;15(7):2410-2424.
doi: 10.1039/d3sc05353a. eCollection 2024 Feb 14.

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

Yunsie Chung  1 , William H Green  1
Affiliations

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

Yunsie Chung et al. Chem Sci. .

Abstract

Fast and accurate prediction of solvent effects on reaction rates are crucial for kinetic modeling, chemical process design, and high-throughput solvent screening. Despite the recent advance in machine learning, a scarcity of reliable data has hindered the development of predictive models that are generalizable for diverse reactions and solvents. In this work, we generate a large set of data with the COSMO-RS method for over 28 000 neutral reactions and 295 solvents and train a machine learning model to predict the solvation free energy and solvation enthalpy of activation (ΔΔG‡solv, ΔΔH‡solv) for a solution phase reaction. On unseen reactions, the model achieves mean absolute errors of 0.71 and 1.03 kcal mol-1 for ΔΔG‡solv and ΔΔH‡solv, respectively, relative to the COSMO-RS calculations. The model also provides reliable predictions of relative rate constants within a factor of 4 when tested on experimental data. The presented model can provide nearly instantaneous predictions of kinetic solvent effects or relative rate constants for a broad range of neutral closed-shell or free radical reactions and solvents only based on atom-mapped reaction SMILES and solvent SMILES strings.

PubMed Disclaimer

Conflict of interest statement

There are no conflicts to declare.

Figures

Fig. 1
Fig. 1. Potential energy diagram of a reaction in a gas phase and a solution phase.
Fig. 2
Fig. 2. Schematic of a machine learning model architecture. The model takes an atom-mapped reaction SMILES and a solvent SMILES as inputs.
Fig. 3
Fig. 3. Pre-trained model results on the reaction split and solvent split test sets. (a) RMSE vs. the data set size for the model trained with the RP-solv feature. (b) RMSE error of different additional features for the model trained with the 500k data set. The chosen data set size and feature are marked with dashed vertical and horizontal lines, respectively. The error bars indicate the standard deviation between five folds.
Fig. 4
Fig. 4. Model MAE vs. the number of fine-tuning epochs. The model is trained with no additional feature. The error on the pre-training set is evaluated on the reaction split test set, and the error on the fine-tuning set is evaluated on the random split validation set.
Fig. 5
Fig. 5. Parity plots of the predicted vs. experimental krel values. (a) The pre-trained model with no additional feature. (b) The pre-trained model with the RP-solv feature. (c) The fine-tuned model with no additional feature. (d) The fine-tuned model with the RP-solv feature. The errors are reported in both log10(krel) and ΔG‡rel units, and R2 represents the coefficient of determination.
Fig. 6
Fig. 6. Parity plots and histograms of the ΔΔG‡solv errors on the 5-fold pre-training set reaction split. (a) The pre-trained model with the RP-solv feature. (b) The fine-tuned model with no additional feature. The MAE and RMSE are in kcal mol−1. The numbers of reactions, solvents, total data points found in the test set are provided. The top and right subfigures on the parity plots show the distribution of computed and predicted values, and the colorbars display the scale of the 2D kernel density estimate plots.
Fig. 7
Fig. 7. The results of the fine-tuned model with no additional feature on the pre-training test set reaction split. (a) Distribution of the ΔΔG‡solv MAE categorized by the number of training reaction data found in each reaction type. (b) Distribution of the ΔΔG‡solv MAE for the 10 most frequent reaction types. The reaction type is specified by the bond changes. For example, +C–H, –C–H, –C–C indicates that one carbon–hydrogen bond is formed, one carbon–hydrogen bond is broken, and one carbon–carbon bond is broken. Outliers are not shown in the plots.
Fig. 8
Fig. 8. Distribution of the ΔΔG‡solv MAE categorized by the number of rotatable bonds found in the reactant(s) of each reaction (outliers not shown). The N indicates the number of reactions in each distribution. The fine-tuned model without additional feature tested on the pre-training set reaction split.
See this image and copyright information in PMC

References

    1. Vermeire F. H. Aravindakshan S. U. Jocher A. Liu M. Chu T.-C. Hawtof R. E. Van de Vijver R. Prendergast M. B. Van Geem K. M. Green W. H. Detailed Kinetic Modeling for the Pyrolysis of a Jet A Surrogate. Energy Fuels. 2022;36:1304–1315. doi: 10.1021/acs.energyfuels.1c03315. - DOI
    1. Payne A. M. Spiekermann K. A. Green W. H. Detailed Reaction Mechanism for 350–400 °C Pyrolysis of an Alkane, Aromatic, and Long-Chain Alkylaromatic Mixture. Energy Fuels. 2022;36:1635–1646. doi: 10.1021/acs.energyfuels.1c03345. - DOI
    1. Edeleva M. Van Steenberge P. H. Sabbe M. K. D’hooge D. R. Connecting gas-phase computational chemistry to condensed phase kinetic modeling: The state-of-the-art. Polymers. 2021;13:1–39. doi: 10.3390/polym13183027. - DOI - PMC - PubMed
    1. Chatelain K. Nicolle A. Ben Amara A. Catoire L. Starck L. Wide Range Experimental and Kinetic Modeling Study of Chain Length Impact on n-Alkanes Autoxidation. Energy Fuels. 2016;30:1294–1303.
    1. Wu H. Grinberg Dana A. Ranasinghe D. S. Pickard F. C. Wood G. P. F. Zelesky T. Sluggett G. W. Mustakis J. Green W. H. Kinetic Modeling of API Oxidation: (2) Imipramine Stress Testing. Mol. Pharm. 2022;19:1526–1539. doi: 10.1021/acs.molpharmaceut.2c00043. - DOI - PubMed