Abstract: Octyl alcohol/air partition (K_OA) is an important parameter to describe distribution of organic chemicals in the air and environment and critical to evaluation of distribution, migration, transformation and ecological effects of chemicals. A total of 309 chemicals were structurally optimized with quantum chemistry methods and screened with the genetic algorithm for optimal descriptors. A quantitative structure-activity relationship (QSAR) model of log K_OA was developed using the multiple linear regression (MLR) and neural network algorithm. The developed QSAR model indicates that the three Dragon descriptors (Nsk, Mor12u and HDon) were the main factors affecting Octyl alcohol/air partition coefficient of chemicals. The resultof fitting with the QSAR model shows that the R² and SE (standard error)of the multiple linear regressions model was 0.911 and 0.880, respectively, while the R² and RMSE(root mean square error) of the neural network model was 0.839 and 0.830, respectively. Application domain of the QSAR model was evaluated with the leverage method. The evaluation shows that the model is quite high in goodness-of-fit, robustness and predictive ability. Therefore, the QSAR model can be used to make up the missing data of lg K_OA, thus lowering the cost of testing and uncertainty of data evaluation.

Key words: qrganic chemicals, octyl alcohol/air partition coefficient, quantitative structure - activity relationship (QSAR)