Prediction of Electrophoretic Mobilities of Organic Acids Using Artificial Neural Networks with Three Different Training Functions

Prediction of Biodegradability for Polycyclic Aromatic Hydrocarbons Using Various In Silico Modeling Methods

Polycyclic aromatic hydrocarbons (PAHs) have attracted great concern as global environmental pollutants. In this work, the quantitative structure-biodegradability relationship (QSBR) study has been done to predict the biodegradability of PAHs and develop the correlation between the biodegradability and the molecular structures. The structural chemistry and quantum chemistry descriptors were used to represent molecular structures. Three in silico modeling methods, i.e., multiple linear regression (MLR), radial basis function neural network, and back-propagation artificial neural network (BPANN), are utilized to construct the linear and nonlinear prediction models and provide some insights into the structural characteristics affecting the biodegradability of PAHs. The stability of these QSBR models was tested by leave-one-out cross-validation, and the cross-validated correlation coefficients (q²) were 0.6109, 0.6887, and 0.6586, respectively. The correlation coefficients (R²) of the three models for the training set were 0.7811, 0.8883, and 0.9667, respectively. The comparison of the three models showed that the BPANN model produced a statistically more significant model than the other two models. On the basis of molecular structure, the dominant molecular structure descriptor affecting biodegradability of PAHs were analyzed and discussed.