MBind: A Web-Based Platform for Metalloprotein and Nonmetalloprotein Docking with Optional ML Docking Integration

Metalloprotein docking is difficult because it requires precise preparation of the grid box, charge assignment, and metal coordination geometry. Many current graphical user interfaces (GUIs) do not support these steps well. Here, we present MBind v2026, a web-based GUI that combines AutoDock Vina, AutoGrid4 and AutoDock4 into a single workflow for standard docking and metalloprotein docking. The interface also has the optional machine learning (ML) docking program GNINA v1.0. The main validated metalloprotein workflow in MBind is based on AutoDock4Zn parameterization for zinc, while preliminary workflows for magnesium, iron, and copper are also included. MBind was benchmarked on eight zinc metalloproteins and eight nonmetalloprotein targets. Its performance was compared with ML pose prediction methods (GNINA v1.0, EquiBind v2026, TankBind v2026, and GAABind v2026), cofolding models (Boltz-2 v2026, and AlphaFold 3 v2026), ML affinity prediction tools (StructureNet v2026, GNNSeq v2026, and PLAIG v2026), and non-ML docking tools (SwissDock v2026, CB-Dock2 v2026, Webina v2026, 1-ClickDock v2026, and MolModa v1.01). Pose accuracy was measured by symmetry-aware ligand RMSD using DockRMSD v1.1 under redocking conditions with co-crystal-defined binding sites. Binding energy trends were evaluated by comparing docking scores with IC50-derived ΔG values using mean absolute error. On the zinc metalloprotein benchmark set, MBind produced a mean RMSD of 0.49 Å. On the nonmetalloprotein set, the mean RMSD was 0.62 Å. These results show that MBind can execute zinc metalloprotein and nonmetalloprotein docking workflows through a web-based interface, while the preliminary Mg, Fe, and Cu workflows require additional validation. The MBind GUI web-based platform is publicly available.