QSAR modelling of water quality indices of alkylphenol pollutants

The aim of this study was to determine the degradability of 26 Alkylphenols (APs) by Chemical Oxygen Demand (COD) and/or 5-day Biochemical Oxygen Demand (BOD(5)), and to describe these data from Quantitative Structure-activity Relationships (QSARs). Statistical analysis techniques, such as Multiple Linear Regression (MLR), Principal Component Regression (PCR), Partial Least-Squares (PLS) Regression and Neural Network (NN) were carried out to calibrate and validate four-descriptor QSAR models using two different types of descriptor sets. Stable MLR-QSAR models using Leave-One-Out (LOO) were obtained with high predictability performance: r(2) = 0.924, Q(2)(cv) =0.854 for log (1/BOD) model on 24 APs and r(2) = 0.888, Q(2)(cv) = 0.818 for log (1/COD) on all the studied APs. The MLR models, built with four Dragon descriptors selected by Genetic Algorithm (GA), presented the following performances on 24 APs: r(2) = 0.889, Q(2)(cv) = 0.848 for log (1/BOD(5)) and r(2) = 0.885, Q(2)(cv) = 0.834 for log (1/COD) on 26 compounds. From these results, it is expected that the QSAR models generated could be successfully expanded to predict the biological and chemical activities of structurally diverse AP compounds.