Friday, November 6, 2009: 9:40 AM
Santa Fe (Camino Real Hotel)
Chlorine is an important disinfectant in drinking water treatment systems. However, prediction of chlorine demand from natural organic matter (NOM) during disinfection has been a challenge to researchers. In this study we report a QSPR model for predicting chlorine demand using model compounds. We used 201 model compounds with chlorine demand ranged from 0.1 to 13.4 mM chlorine per mM of compound. The compounds were divided into calibration set (N = 109), cross validation set (N = 59) and external validation set (N = 42). Multiple linear regression was used to calibrate the model using Minitab 15 and the final QSPR model has nine constitutional descriptors. The statistics of fit for the model were R2 = 0.87, q2 = 0.87, and RMSE = 1.23 mM/mM. Predictive power of the model was assessed by Leave-Many-Out cross validation which gave R2 = 0.87 and q2 = 0.87 and RMSE = 1.16 mM/mM and external validation gave R2 = 92, q2 = 0.90 and RMSE = 1.05 mM/mM. These statistics of cross validation are close to each other, and both meet criteria of model predictability which indicates a robust model. Applicability domain of the model was also assessed and leverage indicates that 5 out 42 validation compounds were over-extrapolated (h ≥ 2.8). The work ahead is to integrate the QSPR with AlphaStep model of natural organic matter to predict chlorine demand for surface water.