Predict Protein Binding Sites Effortlessly with AllMetal3D on Vecura

What is AllMetal3D?

AllMetal3D is a specialized computational tool composed of two 3D convolutional neural networks—AllMetal3D and Water3D—designed to predict the binding sites of metal ions and water molecules within protein structures. By analyzing the local chemical environment of each residue, the tool generates probability density grids that identify potential binding locations. It is particularly valuable for structural biologists and researchers who need to annotate protein structures, guide mutagenesis experiments, or interpret complex electron density maps without requiring intensive manual inspection.

What can users do with AllMetal3D on Vecura?

With AllMetal3D on Vecura, users can:

• Predict metal-ion and water binding sites directly from PDB or mmCIF files.
• Customize inference using different modes, such as 'fast' for high-efficiency processing or 'site' for targeted analysis of specific residues.
• Export results as PDB probe files and Gaussian CUBE files for seamless integration with visualization software like UCSF ChimeraX and PyMOL.
• Obtain detailed JSON reports containing per-site probability vectors for metal identity and coordination geometry, enabling data-driven site filtering.

What the output means

The output provides a combination of machine-readable data and structural files. Users receive PDB files mapping site centroids and Gaussian CUBE files that represent the probability density, allowing for direct visual inspection alongside the original protein structure. The included JSON data provides granular insights into the confidence level, identity, and coordination geometry of predicted sites.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

Understanding where metal ions and water molecules bind is essential for deciphering protein function, stability, and enzymatic activity. Traditional experimental methods to identify these sites can be time-consuming and sometimes ambiguous, especially when dealing with low-resolution structural data or flexible protein regions.

AllMetal3D offers an automated, high-throughput solution to these challenges by leveraging 3D convolutional neural networks to predict binding propensities based on the protein's local electrostatic and steric environment. This enables researchers to formulate hypotheses about metal-protein interactions faster and with greater confidence.

• Developed by: LCBC Laboratory, EPFL
• Source: GitHub, BioRxiv
• Reference: Duerr and Roethlisberger (BioRxiv, 2024)