A new quantitative structure–property relationship model to predict bioconcentration factors of polychlorinated biphenyls (PCBs) in fishes using E-state index and topological descriptors

A quantitative structure–property relationship (QSPR) study for predicting the logarithm of bioconcentration factors (Log BCF) of polychlorinated biphenyls (PCBs) is presented in this work. For this, the descriptors were obtained using only the Simplified Molecular Input Line Entry System (SMILES) strings in the free web server Parameter Client. The model was built using the Partial Least Squares (PLS) regression method. The best model presented five descriptors (one E-state index and four topological descriptors) and a high quality for fit, internal, and external predictions. The leave-N-out (LNO) cross validation and the y-randomization test showed the model is robust and has no shown chance correlation. With a second test set, the model was compared to other models and presented a root mean square error (RMSE) very close to the best model. The mechanistic interpretation was corroborated by other works in the literature and by the descriptors' theory. Thus, the results meet the five Organization for Economic Co-operation and Development (OECD) principles for validation of QSA(P)R models, and it is expected the model can effectively predict the BCF values in fishes of the PCB congeners without highly reliable experimental BCF.

Graphical AbstractA multivariate quantitative structure-property relationship (QSPR) study for predicting the Log BCF of polychlorinated biphenyls (PCBs) was carried out using descriptors calculated by means of SMILES strings, Partial Least Squares (PLS) as the calibration method, and Ordered Predictors Selection (OPS) for variable selection.Figure optionsDownload as PowerPoint slideHighlights
► There are few QSPR models for the prediction of bioconcentration factor (BCF) for PCBs.
► We used a data set with 62 PCBs to develop a new model for the prediction of Log BCF.
► Ordered Predictors Selection (OPS) was used for variable selection and the descriptors were obtained with SMILES.
► The model obtained is in accordance to the OECD principles.
► Descriptors are related to the Log BCF and supported by the literature.