Comparative Study of Depuration Rate Prediction against Mussel (Elliptio complanata) using Different Chemometric Approaches

Abstract

Different chemometric approaches were applied to a heterogeneous dataset of persistent organic pollutants(POPs), which included polybrominated diphenyl ethers(PBDEs), polychlorinated biphenyls(PCBs) and polycyclic aromatic hydrocarbons(PAHs) with associated depuration rate constant in Mussel (Elliptio complanata), to develop robust quantitative structure-activity relationship(QSAR) models. These models were further validated for statistical significance and predictive ability by internal and external validation. Out of various methods available, genetic algorithm and principal component analysis (PCA) approaches were used to identify relevant molecular descriptors from a large descriptor pool that exhibited a strong correlation with the depuration rate constant values of the diverse dataset. Then, multiple linear regression(MLR) and artificial neural network (ANN) methods were applied to the selected descriptors to create good predictive models. Statistical comparison of 3 hybrid approaches namely, GA-MLR, GA-ANN and PCA-ANN have shown that the genetic algorithm coupled with ANN model is superior to the other 2 models (R2train= 0.961, R2test= 0.947, mapetest= 7.939 and rmsetest=0.128). The applicability domain of the selected models was analyzed using the Euclidean distance and leverage approach signifies that all test set compounds fall within the applicability domain of the developed regression-based models.
 HIGHLIGHTS
 
 Evaluation of toxicokinetic parameters in aquatic ecosystem from the molecular structure satisfies the growing demand of theoretical methods for sustainable chemistry
 Three different chemometric approaches namely GA-MLR, GA-ANN and PCA-ANN are applied to establish quantitative-structure-activity relationships for the prediction of depuration rate constant values (logkd) of a dataset containing POPs in mussel Elliptio complanata
 OECD guidelines are used to evaluate and validate the presented models