Quantitative structure-property relationships for predicting chlorine demand and disinfection byproducts formation in drinking water

Models are important tools for designing or redesigning water treatment processes and technologies to minimize disinfection byproducts (DBPs) formation without compromising disinfection efficiency. Empirical models, which are the most common, are based on bulk water quality parameters that vary with time and space. These parameters may not always have linear relationships with chlorine demand and DBPs formation which make structure-based models more attractive to study. In this dissertation, Quantitative Structure-Property Relationship (QSPR) models which make use of structural properties of individual molecules were developed using experimental data obtained from the literature. The amounts are reported in moles of chlorine (HOCl) consumed or DBP formed per mole of a compound (Cp). The QSPRs were derived by multiple linear regression of chlorine demand or DBPs on a set of significant constitutional descriptors. The QSPRs were also tested for predictive power using cross validation and external validation for which the criteria were: Rc2 > 0.6, q2 > 0.5, 0.85 ≤ k ≤ 1.15 and Rt = (Ri2-Ro2)/Ri2 < 0.1. The eight descriptor QSPR for HOCl demand had good statistics of fit (Rc2 = 0.86 and SDE = 1.24 mol-HOCl/mol-Cp, N = 159) and also showed high predictive power on cross validation data (q2LMO = 0.86, RMSELMO = 1.21 mol-Cl2/mol-Cp) and external validation data (q2ext = 0.88, RMSELMO = 1.17 mol-HOCl/mol-Cp). The QSPR also met all the criteria for QSPR predictive power and was robust. This model was integrated with AlphaStep model of natural organic matter (NOM) so as to estimate chlorine demand of surface waters. The predicted chlorine demand was 27.55 μmol-HOCl/mg-C which is comparable to 27-33 μmol-HOCl/mg-C reported for surface waters. The 4 descriptor QSPR for total organic halide (TOX) formation had Rc2 = 0.72 and SDE = 0.43 mol-Cl/mol-Cp. The Leave-One-Out validation of the QSPR (q2LOO = 0.60, RMSE = 0.5 mol-Cl/mol-Cp, N = 49) and external validation (q2Ext = 0.67, RMSE = 0.48 mol-Cl/mol-Cp, N = 12). These statistics showed that the QSPR had high predictive power and also was robust. Results from integration of the QSPR with AlphaStep predicted TOX in surface water to be 183.6 μmol-Cl/mg-C which comparable 170-298 μg-Cl/mol-Cp for the experimental TOX formation measured for whole dissolved organic matter. Trichloromethane (TCM) and trichloroacetic (TCAA) were the two specific DBPs studied. The QSPR for TCM formation had three descriptors and statistics of fit were Rc2 = 0.97 and SDE = 0.08 mol-TCM/mol-Cp and was validated by LMO data and external data. The results showed that LMO cross validation (q2LMO = 0.94, RMSE = 0.09 mol-TCM/mol-Cp, N = 90) and external validation (q2Ext = 0.94, RMSE = 0.08 mol-TCM/mol-Cp, N = 27) met criteria of predictive power and was therefore robust. The model prediction of 0.33 mol-TCM/mol-Cp was higher than 0.13 mol-TCM/mol-Cp observed for tannic acid. The QSPRs for predicting TCAA formation were developed but none of them met all the criteria for predictive power and were therefore not robust. The relationship between predicted TCAA and experimental data was too weak to be useful. This implies that TCAA formation has insignificant linear relationship with constitutional descriptors and it may better be predicted by QSPRs derived from non-linear algorithms. A major drawback of the constitutional descriptors is that they cannot explain electronic or steric effects. It is not easy to explain the differences in electron density and steric effects when same number of substituents occupy different position relative each other in aromatic ring (e.g., catechol vs. quinol). Use of geometrical descriptors (e.g., molecular volume, solvent accessible area), quantum-chemical descriptors (e.g., dipole moment, polarizability) or electrostatic descriptors (e.g., partial charge, polarity index) is recommended.ChemistryDoctoralUniversity of New Mexico. Dept. of ChemistryCabaniss, StephenWang, WeiEvans, DeborahSchuler, Andre