Wastewater often contains toxic heavy metals (HMs) that can harm aquatic ecosystems if insufficiently treated. One promising approach is adsorption using metal-oxide engineered nanomaterials (ENMs). ENMs are dosed into water to adsorb target HMs, after which ENM–HM complexes are removed by standard separation. A key limitation is that not all ENM–HM combinations achieve high removal efficiencies, underscoring the need for tools that guide the selective choice and design of ENMs for specific HMs. We aimed to identify the mechanistic atomic factors that affect the ability of a given metal-oxide ENM to remove a specific HM, using atomic- and elemental-level descriptors. We assembled a dataset from a previously published study, which contains the removal rates of 11 HMs by 3 different ENMs (α-Fe₂O₃, CuO, and ZnO), yielding 33 ENM–HM combinations [1]. We used the atomic properties for HMs and descriptors of ENMs (calculated using“Elemental Descriptor Calculator”). We combined these descriptors with a gradient-boosted artificial intelligence classifier (XGBoost) trained on an operational label where the positive class denotes complete HM removal (>99% removal). The model achieved an Area Under the Curve metric of 1 on six held-out records, indicating promising but preliminary predictive performance. Importantly, the trained model highlighted the atomic radius of HM and the sum of ionization potentials of constituent elements in the ENM formula as the most important properties for the HM adsorption by ENMs. Mechanistically, the prominence of the HM radius and the ENM’s summed ionization potentials indicates that adsorption is jointly controlled by size-dependent hydration and steric access to inner-sphere sites. Overall, these findings uncover important mechanistic insights about ENM-HM interaction at the atomic level, allowing for better selectivity and ENM design, in line with the IOCAT objectives.

1. Bouafia et al., Sci Rep 13, 5637 (2023). https://doi.org/10.1038/s41598-023-32938-1