QSAR enabled predictions in water treatment: from data to mechanisms and vice-versa

The efficiency of water treatment systems to remove emerging (chemical) substances is often
unknown. Consequently, the prediction of the removal of contaminants in the treatment and supply
chain of drinking water is of great interest. By collecting and processing existing chemical properties of contaminants, QSARs (quantitative structure-activity relationships) for typical removal parameters can be constructed. Depending on the definition of the predicted endpoint, QSARs (1) are embedded in a process model suite, where they serve to predict a model parameter and the total, hybrid model predicts a removal rate or (2) are used to directly predict e.g., the removal rate, or a rejection coefficient for membrane systems. The different types of resulting prediction models, ranging from mechanistic (causal) to empirical (data-based), allow for hypothesis testing of current physico-chemical mechanisms and interactions between the contaminant, the type of water and the materials or energy (e.g. UV light) of the removal barrier. Two case studies illustrate this viewpoint and also pinpoint that, firstly, QSAR development, validation and residual analysis stress the linkage between the QSAR endpoints and process model predictions, and secondly, they lay bare the need to share data, algorithms and models.