Quantitative structure-property relationship for predicting chlorine demand by organic molecules

Conventional methods for predicting chlorine demand (HOCl(dem)) due to dissolved organic matter (DOM) are based on bulk water quality parameters and ignore structural features of individual molecules that may better indicate reactivity toward the disinfectant. The Quantitative Structure-Property Relationship (QSPR) modeling approach can account for structural properties of individual molecules. Here we report a QSPR for HOCl(dem) based on eight constitutional descriptors. Model compounds with HOCl(dem) ranging from 0.1 to 13.4 mol chlorine per mole compound were divided into a calibration and cross-validation data set (N = 159) and an external validation set (N = 42). The QSPR was calibrated using multiple linear regression in a 5-way leave-many-out approach and has average R(2) = 0.86 and standard error of regression (StdE(reg)) = 1.24 mol HOCl per mole compound and p < 0.05. Internal cross-validation has average q(2) = 0.85 and the external validation has q(2) = 0.88, indicating a robust model. The leverage of 7 of 42 compounds in the external validation data set exceeded the critical value, suggesting that these compounds may be overextrapolated. However, root-mean-square error of prediction in the external validation was 1.17 mol HOCl per mole compound, and all compounds were predicted with +/-2.5 standardized residuals (Sresid). Application of the QSPR to model structures of NOM predicts HOCl(dem) comparable to reported measurements from natural water treatment.