Answers are grounded in this paper and cite their sources.
Accurate prediction of ligand binding modes in non-heme metalloproteins remains challenging because conventional docking methods often fail to correctly identify poses with accurate metal-coordination geometries. Here, we introduce CHELATE, a machine learning-based post-docking filter that predicts the root-mean-square deviation (RMSD) of ligand metal-binding groups (MBGs) to filter out inaccurate poses and prioritise those with native-like metal coordination. We curated a dataset of over 50,000 metalloprotein–ligand systems spanning more than 20 metal ions, split based on protein-ligand similarity to enable evaluation on structurally distinct systems. Cross-docking benchmarks showed that, although existing docking methods often sample poses with accurate metal-binding group placement, they fail to rank them correctly. We quantify this using the MBG success rate, defined as the fraction of top-ranked poses with an MBG RMSD ≤ 1 Å relative to the crystal structure. Applying CHELATE as a post-docking filter improved MBG success rates from a baseline of 18% across all systems to 32%, 34%, and 40% for AutoDock Vina, smina, and GNINA, respectively, for systems for which at least one docking pose passed the filter. Feature importance analysis indicates that CHELATE predictions are informed by chemically meaningful features associated with metal coordination. Representative case studies illustrate the scope and limitations of CHELATE. Its low computational cost and compatibility with existing docking workflows make it a promising filtering strategy for a broad range of structure prediction methods, including protein–ligand co-folding models.
No takes yet. Share an insight, caveat, or question.
Görlich et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: