A rapid technique for approximating the zeta potential of metal oxide nanoparticles based on pH measurement and machine learning nano-QSPR models

It is essential to adopt a systematic approach for assessing the risk associated with the deposition and dispersion of engineered nanomaterials (ENMs) in the environment. An adequately designed risk management system is crucial to protect human health and life. The comprehensive characterization of the properties of ENMs is a challenging undertaking due to the various shapes, sizes, and types of nanomaterials currently available. The application of machine learning (ML) methods represents a potential solution that can assist in this process. From the environmental perspective, one of the most critical characteristics of ENMs is zeta potential (<i>ζ</i>). Our research findings have led to the development of a predictive model that enables the estimation of <i>ζ</i> for nano-MeOx, utilising experimentally determined pH values and simple descriptors of the nano-structure. We have projected the optimal methodology using an algebraic approach of integrating nano-QSPR (Quantitative Structure-Property Relationship) models to obtain the best possible explanation, avoid losing important information, and produce a robust consensus model for five selected MeOx (Al₂O₃, CeO₂, Fe₂O₃, MnO₂, ZnO). The developed model demonstrates a high predictive ability (<i>Q</i>_F1-3 > 0.897) and goodness-of-fit (<i>R</i>² = 0.912). A distinctive attribute of the model we have devised is its capacity to rapidly approximate <i>ζ</i> through the utilization of an environmental variable (pH) in conjunction with readily calculable structural descriptors.