Development of QSBR models for anoxic biodegradability of polycyclic aromatic hydrocarbons by using SMLR and BP-ANN

In the present study, quantitative structure-biodegradability relationship (QSBR) models were developed for the anoxic Andrews model parameters of polycyclic aromatic hydrocarbons (PAHs). The molecular geometries of 20 PAHs were studied using density functional theory. Stepwise multiple linear regression (SMLR) and backpropagation artificial neural network (BP-ANN) methods were applied to establish the QSBR model between the anoxic biodegradability (qmax) and 5 molecular descriptors. Results of the regression analysis indicated that both the accuracy and predictive ability of the BP-ANN model were better than those of SMLR model. After analyzing the sensitivity of variables, the key molecular structure descriptor influencing anoxic biodegradability of PAHs were screened to be EHOMO and Freq. The present study demonstrates the value of QSBR not only as a predictive tool but also as a framework for understanding the mechanisms governing biodegradation at the molecular level.