Molecular electronegativity distance vector model for the Prediction of bioconcentration factors in fish

Molecular electronegativity distance vector (MEDV) derived directly from the molecular topological structures was used to describe the structures of 122 nonionic organic compounds (NOCs) and a quantitative relationship between the MEDV descriptors and the bioconcentration factors (BCF) of NOCs in fish was developed using the variable selection and modeling based on prediction (VSMP). It was found that some main structural factors influencing the BCFs of NOCs are the substructures expressed by four atomic types of nos. 2, 3, 5, and 13, i.e., atom groups −CH2− or =CH−, −CH< or =C<, −NH2, and −Cl or −Br where the former two groups exist in the molecular skeleton of NOC and the latter three groups are related closely to the substituting groups on a benzene ring. The best 5-variable model, with the correlation coefficient (r2) of 0.9500 and the leave-one-out cross-validation correlation coefficient (q2) of 0.9428, was built by multiple linear regressions, which shows a good estimation ability and stability. A predictive power for the external samples was tested by the model from the training set of 80 NOCs and the predictive correlation coefficient (u2) for the 42 external samples in the test set was 0.9028.