Multivariate analysis about the influence of solubility and partition molecular descriptors on LogKoc for substituted diphenyl ethers

Substituted diphenyl ethers are compounds used in pesticides, pharmaceuticals, cosmetics, and other industrial applications. They exhibit high chemical stability and environmental persistence, raising concerns about toxicity and environmental fate. The soil organic carbon–water partition coefficient (Log K oc ) is a key parameter for evaluating this property. It is commonly modeled using descriptors related to hydrophobicity (Log P ) and aqueous solubility (Log S ). The predictive performance of such models may depend strongly on the computational algorithms used to derive these descriptors. In this study, a dataset comprising 59 substituted diphenyl ethers with previously determined experimental Log K ow values was evaluated using 11 Log P and 4 Log S algorithms via univariate regressions, with rigorous validation criteria applied. Only the MLog P algorithm satisfied all statistical criteria. Subsequently, multivariate models based on partial least squares (PLS) regression were developed and assessed using LOO and LNO cross-validation, y -randomization tests, and applicability domain analysis. In addition, the hybrid descriptor Log S p was investigated both in the original formulation and in a recalculated version derived from algorithms optimized for the present dataset. Consistent with prior studies, the results indicate that different combinations of Log P and Log S algorithms can directly affect the quality of Log K oc predictive models, resulting in either good or poor performance. Moreover, the model incorporating the hybrid Log S p descriptor, as originally proposed and now applied to a structurally homogeneous dataset, yielded the best QSPR model in terms of both internal and external quality, highlighting its potential for predicting this important parameter related to the toxicity and environmental fate of chemical compounds.