Use AllMetal3D Online
Commercially Available AllMetal3D No-Code Web Server
AllMetal3D: Predict Metal Ion Location, Identity and Coordination Geometry in Proteins
AllMetal3D is a deep-learning tool that jointly predicts the location, identity and coordination geometry of common metal ions in protein structures. It was developed by Simon L. Dürr and Ursula Rothlisberger at LCBC, EPFL, and described in a 2025 bioRxiv preprint. It extends Metal3D, which was originally trained on zinc sites, to alkali, alkaline-earth and transition metals.
Metal ions are common protein cofactors, and Metal3D, the starting point for this work, was trained on zinc sites only. The authors also report data bias, weak negative-site prediction and limited selectivity in several comparison tools.
Property | Detail |
|---|---|
Developers | Dürr and Rothlisberger, LCBC, ISIC SB, EPFL |
Input | Protein structure, typically a biological assembly; no metal identity or count needed |
Outputs | Site coordinates with probability, identity probabilities, geometry probabilities |
Metal classes | MG, CA, NA, K, ZN, FE, FE2, CU, CO, CU1, MN |
Access | pip package, ChimeraX plugin, Gradio web app |
How AllMetal3D Predicts Metal Binding Sites
Locate sites: AllMetal3D_loc, a retrained Metal3D model, predicts a combined metal-density channel on 32x32x32 voxel grids.
Fingerprint each site: AllMetal3D_base produces a 1280-length fingerprint for each predicted site.
Classify: separate heads predict identity (including a NoMetal class) and geometry (six geometry classes plus NoMetal).
Apply thresholds: the location threshold is p = 0.65, with density peaks kept above p = 0.25.
Performance of AllMetal3D on Location, Identity and Geometry
Task | Reported result |
|---|---|
Zn2+ location | Mean probability 0.90 vs 0.97 for Metal3D |
Spatial deviation | Median 0.5 A for Ca2+, Ni2+, Cu2+, Zn2+; 0.6 A for Mg2+, Na+, K+ |
Non-metalloprotein set (528) | 2 pockets flagged above p = 0.5 |
Identity, full 9-class model | 46% accuracy, 80% top-3 |
Example
For carbonic anhydrase (PDB 2CBA), the tool identified Zn2+, with tetrahedral geometry at 30% and octahedral at 19%.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with AllMetal3D on Tamarind Bio
Locating missing ions: find probable metal sites in predicted or apo structures.
Metalloenzyme design: check whether a designed site is likely to bind a metal.
Structure validation: test experimental metal assignments, including negative sites.
Geometry assessment: judge whether a site looks tetrahedral or octahedral.
How to Use AllMetal3D on Tamarind Bio
Log in and open the tool: sign in at tamarind.bio and select AllMetal3D.
Provide a structure: upload a PDB file, ideally a biological assembly.
Set parameters: keep the default probability thresholds unless exposed.
Run: submit the job.
Download outputs: collect site coordinates and probabilities.
Interpret: read identity and geometry probabilities per site, treating NoMetal as a possible negative.
Use downstream: place ions in models or guide mutagenesis.
Things to Keep in Mind
Performance is weaker for Na+, K+ and Mg2+ than for transition metals.
Geometry prediction works mainly for tetrahedral and octahedral sites.
Identity uses merged classes; fine transition-metal selectivity was not achieved.
Positions carry about 0.5 A imprecision.
Proteins containing DNA were excluded.
Source: Dürr SL, Rothlisberger U, "AllMetal3D: joint prediction of localization, identity and coordination geometry of common metal ions in proteins", bioRxiv preprint (2025), DOI 10.1101/2025.02.05.636627. Code: https://github.com/lcbc-epfl/allmetal3d.