Predicting unknown binding sites for transition-metal-based compounds in proteins
Andrea Levy, Ursula Rothlisberger
Abstract
Transition-metal-based compounds are promising therapeutic agents, particularly in cancer treatment. However, predicting the binding sites of such compounds remains a major challenge. In this work, we investigate the applicability of two tools, Metal3D and Metal1D, for this purpose. Although originally trained to predict zinc ion binding sites only, both predictors correctly identify at least one of the experimentally observed binding sites for transition metal complexes in each of the apo protein structures tested.
Introduction
Compounds based on transition metals (TMs) are promising candidates for biomedical applications, as the combination of organic ligands with TM centers enhances key properties such as stability, solubility, and bioavailability [1,2]. However, some TMs pose considerable health hazards, and TM-based drugs often show severe side effects. This is the case for cisplatin, which, despite its several toxic side effects, remains a chemotherapeutic workhorse for various tumors [3].
Methods
Metal3D and Metal1D predictors
A detailed description of Metal3D and Metal1D can be found in the original publication by Dürr et al. [12] In the same work, the dataset composition is analysed in detail, including the distribution of coordination motifs. Here, we provide a short overview of the main features of the two predictors and highlight their differences and similarities. A schematic representation of their training and inference mode to predict metal ion positions is provided in Fig 1.
Results and discussion
Differences in data availability
In the case of Metal1D, we generated two additional probability maps to compare them with the results from the original one based on zinc, using PDB structures containing Pt or Ru ions, as well as complexes containing such TMs. The generation of the probability map is analogous to the one for zinc described in Ref. [12], and is based on the LINK records in the PDB structures. In the case of Pt, fewer than 250 crystal structures in the PDB database contain a Pt-based compound covalently bound to a biomolecule.
Conclusion
In this work, we explored the applicability of the Metal3D and Metal1D predictors [12] to identify the binding sites of TM-based compounds. Although they were originally designed to predict binding sites of zinc ions, our results demonstrate a surprising degree of predictive power for more complex TM binding sites.
Citation: Levy A, Rothlisberger U (2026) Predicting unknown binding sites for transition-metal-based compounds in proteins. PLoS One 21(6): e0349622. https://doi.org/10.1371/journal.pone.0349622
Editor: Soumendranath Bhakat, AlloTec Bio, UNITED STATES OF AMERICA
Received: February 19, 2026; Accepted: May 1, 2026; Published: June 9, 2026
Copyright: © 2026 Levy, Rothlisberger. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All the predictions for the different protein structures are deposited on Zenodo under https://zenodo.org/records/18416988, including VMD visualization states to reproduce the figures presented in the paper. In the same repository, we also included a Jupyter Notebook to perform all predictions with Metal3D and Metal1D, as well as generate the new probability maps for Metal1D.
Funding: U.R. acknowledges funding from the Swiss National Science Foundation (grant 200020-185092 and 200020-219440). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
