Predicting Biological Removal of Contaminants in Wastewater Treatment: QSBR Modeling

Contaminant fate and transport models are a highly desirable alternative to direct measurement of environmental behavior for the very large number of so-called “emerging contaminants”. In particular, it is desirable to estimate what fraction of emerging contaminant loading is removed by conventional wastewater treatment plant processes, so that the loading of organic wastewater contaminants (OWCs) into the environment can be assessed. In this thesis, we focused on prediction of biodegradation rate constants for removal of OWCs during activated sludge treatment. A Quantitative Structure-Biodegradability Relationship (QSBR) modeling approach was used to predict pseudo first-order biodegradation rate constants (kb) based on molecular descriptors from commercially available computational chemistry software. A training dataset comprising 65 previously measured molecular structures was collected from nine different literature sources. This data was then used to create four QSBR models using varying molecular subsets and their associated descriptors. Internal validation statistics indicate that the overall QSBR model (comprising all compounds in the training set) achieves less predictive ability (R2 =0. 49) than three smaller QSBR models (R2 = 0.97, 0.88, and 0.90) that were created from smaller subsets of the same dataset. External validation was performed via direct measurement of three highly-prescribed, previously unevaluated pharmaceutical OWCs: metformin, benazepril, and warfarin. Their respective kb values were 0.0105, 0.0033, and  0 L/g-h. Of the four QSBR models, the general, all-encompassing model delivered highly accurate predictions for metformin Logkb (measured = -1.98 versus predicted = -2.03) and benazepril Logkb (measured = -2.48 versus predicted = -2.29), while warfarin was best estimated using one of the smaller subset QSBRs. Analysis of external validation results also indicated that the diversity of the molecules comprising each model’s underlying dataset should be used to assess each model’s application domain before the model is used to make predictions for unmeasured compounds. Other results from the QSBR models, including identification of which molecular descriptors are best correlated with biodegradation rate constant, offer new information on particular contaminants of concern in surface and groundwater systems and the nature of WWTP biodegradation reactions. Future work will focus on comparisons between QSBR and other parameter modeling approaches.