QSPR model of Henry’s law constant for a diverse set of organic chemicals based on genetic algorithm-radial basis function network approach

Six quantitative structure–property relationship (QSPR) models for a diverse set of experimental data of Henry’s law constant (H) of organic chemicals under environmental condition (T = 25 °C; water–air system) have been developed based on four different molecular descriptor sets. Three different models based on the descriptors of CODESSA (Comprehensive Descriptors for Structural and Statistical Analysis), Tsar, and Dragon software and a model based on a combined descriptor set from these packages, and in addition from HYBOT software, have been established using the stepwise regression method. The combined descriptors set model gave the best results. Furthermore, a genetic algorithm was used for descriptor selection from a combined set of descriptors, and a radial basis function network was utilized to establish a model with a low root mean square error (RMSE). The results of this study were compared with the well-known bond contribution and group contribution methods. The group contribution method failed to predict Henry’s law constant of 170 from all 940 compounds in the data-set. RMSEs of 0.693, 0.798, and 0.564 were achieved for bond contribution, group contribution and the best QSPR model of this study, respectively, based on logarithm of H. Analysis of different QSPR models showed that hydrogen bonding between the organic solute and water as a solvent has the greatest influence on this partitioning phenomenon.