MetalKB: Predicting Metal Binding Sites on Proteins with a Knowledge-Based Graph Framework

Metal ions play a crucial role in the function, regulation, and stability of proteins. Therefore, accurate prediction of metal ions' binding sites is valuable to reveal the molecular mechanism of related biological processes. Here, we propose MetalKB, a novel knowledge-based framework for predicting the binding sites of metal ions on proteins by using atomic-level statistical potentials and graph-theoretical strategies. Specifically, possible donor atom clusters are first identified using a clique detection algorithm, from which initial metal ion coordinates are generated. These candidate coordinates are then evaluated and locally refined using knowledge-based statistical potentials derived from a protein-metal ion binding database. Redundant predictions are subsequently removed by applying spatial distance thresholds. Evaluations on diverse benchmark data sets provided by Metal3D and TEMSP show that MetalKB demonstrates competitive performance compared with seven representative methods in terms of precision, recall, and F1 score, while exhibiting strong robustness and parameter stability. MetalKB is capable of identifying complex coordination environments, including multinuclear and bridging metal-binding sites, as illustrated in representative structural examples. In addition, it also provides prediction of both metal ion 3D coordinates and residue-level coordinating ligands.