Integrating Experiments and Machine Learning Modeling to Assess the Half-Wave Potentials of Antibiotics

Antibiotics are emerging organic contaminants widely distributed in the environment. Understanding their redox properties could help evaluate their environmental fate and design effective treatment strategies. Here, half-wave potentials (E1/2) offer valuable insights into the redox behavior of compounds, but data availability is limited for many antibiotics. A series of cyclic voltammetry (CV) experiments were conducted to determine E1/2 of 23 antibiotics from 7 groups under three pH conditions. We found that almost all antibiotics underwent irreversible oxidation, with sulfonamides and tetracyclines exhibiting pH-dependent shifts in their E1/2 values, representing Nernstian behavior. Quantitative structure–property relationship (QSPR) models were developed using stepwise multiple linear regression (SW-MLR) and ten machine learning algorithms to investigate the influence of molecular structures and identify the best predictive model. Both the SW-MLR and adaptive boosting (AdaBoost) models demonstrated strong performance, characterized by high goodness-of-fit and predictive abilities. In the SW-MLR, molecular structural connectivity and topological complexity were identified as the most influential features, while charge distribution was found to play a dominant role in capturing nonlinear relationships in the AdaBoost model. This work deepens the understanding of the redox behavior of antibiotics and proposes a novel QSPR model for predicting E1/2. Thereby, the model provides indications for their susceptibility against (bio)electrochemical oxidation, however, these predictions need to be verified in corresponding electrochemical and bioelectrochemical degradation studies.