Predicting aqueous-phase hydroxyl radical reaction kinetics with organic compounds in water, atmosphere, and biological systems

The hydroxyl radical (HO·) acts as a reactive electrophile, playing crucial roles in water, the atmosphere, and biological systems. In this study, an interpretable model that employs six structural descriptors assesses the second-order rate constants (k_HO·) for various classes of organic compounds with diverse functional groups. The training dataset comprises measured k_HO· values at 25°C in the aqueous phase for over 1000 organic compounds including 876 data points of the training set, while the test (101 data points) and validation (110 data points) sets contribute to the development of the new models. The second-order rate constant model is based on the interaction of the HO· radical with different atoms and functional groups. To account for significant deviations in predicted results, an additional variable is introduced into the core model, resulting in an improved version. The calculated data from the enhanced model are compared with the outputs of a combined molecular fingerprint-machine learning complex method, considered the best available general approach. For the test set, the ratios of the maximum absolute error (AE_max), the average absolute error (AAE), R², and root mean square error (RMSE) between the improved model and the comparative method are as follows: 0.767/0.806, 0.213/0.222, 0.7197/0.6716, and 0.263/0.294, respectively. These ratios exhibit consistent trends in the validation set: 0.856/1.11, 0.207/0.226, 0.270/0.302, and 0.7830/0.7346 for AE_max, AAE, RMSE, and R², respectively. The reliability of the improved and interpretable model surpasses that of the comparative complex method when applied to new organic compounds not utilized in deriving the model.