How do water matrices influence QSPR models in wastewater treatment?–A case study on the sonolytic elimination of phenol derivates

As the demand of freshwater increases with simultaneously aggravated climatic challenges, the development of efficient and effective water purification methods is of high importance. Qualitative Structure-Property Relationships (QSPRs) can support this process by calculating a correlation between the molecular structure and the degradability of water pollutants in a defined removal procedure, expressed by the kinetic constant of their removal. This can help to receive more mechanistical interpretation of the underlying process, but also to reduce experimental costs and time. As most QSPR models in wastewater treatment research are based on experimental data using ultrapure water as reaction solutions, it is still unknown to which extent QSPR models for different water matrices differ from each other with regard to selected descriptors and performance. Therefore, in this study the sonolytic degradation of 32 phenol derivates was investigated for three different water matrices (NaCl, Glucose, NaCl+Glucose) and compared to a previous study in ultrapure water. With only very few exceptions, the addition of water additives reduced the degradability of the target analytes. Based on these four datasets, QSPR modelling, respecting all five OECD principles for reliable QSPR models, were performed using numerous internal and external validations as well as statistical quality assurances to ensure good regression abilities as well as stability and predictivity. As the final four models were compared, it was observed that the descriptor selection and model calculation were highly impacted by the water additives. This was also confirmed when the descriptor pools of the best 10 models for each water composition were compared, as the descriptor pools were also highly dissimilar, indicating a shift in structural importance when changing the water composition. It could be shown that water matrices significantly influence the results of QSPR modelling even at very low concentrations of the matrix components.