Prediction of n-octanol/water partition coefficients (Kow) for pesticides using a multiple linear regression-based QSPR model

This study focused on predicting the n-octanol/water partition coefficient (Kow) for a set of 56 pesticides using a quantitative structure-property relationship (QSPR) approach. The analysis used multiple linear regression (MLR) to relate the logarithm of Kow values (log Kow) to molecular descriptors derived from the chemical structures. The dataset was divided into a training set of 42 compounds and a test set of 14 compounds using the Kennard-Stone algorithm, which ensures even coverage of the descriptor space and improves the quality of external validation by reducing sampling bias.

The molecular descriptors were calculated using Dragon software. A genetic algorithm (GA) combined with a variable subset selection (VSS) procedure was employed to identify and retain the most informative descriptors. This approach helped to build a predictive model with reduced complexity and enhanced interpretability.

The model showed strong statistical performance, with R² = 93.22, Q²LOO = 90.89, Q²ext = 92.77, SDEC = 0.450, SDEP = 0.520, SDEPext = 0.546, F = 92.4052, and s = 0.511. Internal cross-validation and external test set validation confirmed the model’s robustness, reliability, and predictive power.

These results support the use of the model as a reliable and practical tool for predicting Kow values of pesticides. It contributes to environmental and chemical risk assessments by enabling better evaluation of the environmental behavior, bioaccumulation, and toxicity of pesticide compounds.