AllMetal3D

AllMetal3D predicts the location, identity, and coordination geometry of biologically relevant metal ions—including alkali ions (sodium, potassium), alkaline earth ions (magnesium, calcium), and transition metal ions (zinc, iron, and others)—in protein structures without requiring preselection of a specific ion or location.

Key Features:

• Joint Metal Prediction: Simultaneously predicts localization, identity, and coordination geometry of common metal ions in proteins.
• Multi-Class Metal Support: Extends beyond zinc to cover all major biological metal classes including alkali, alkaline earth, and transition metal ions.
• Negative Site Prediction: Addresses prediction of negative sites, highlighting shortcomings in existing tools regarding data bias and selectivity.
• Coordination Geometry Classification: Classifies metal coordination geometries, with strongest performance on tetrahedral and octahedral arrangements.
• Generalization: Outperforms the predecessor Metal3D framework for Zn²⁺ location prediction and generalizes well to other metals for location prediction and identity classification.

Scientific Applications:

• Metal Cofactor Annotation: Identifies metal ion binding sites in protein structures for functional characterization.
• Comparative Tool Assessment: Enables benchmarking against existing methods such as MetalSiteHunter, AlphaFold3, MIC, MIB2, MetalHawk, ESMBind, and the original Metal3D framework.
• Structural Biology Research: Supports analysis of metal-dependent protein functions across diverse biological systems.

Methodology:

AllMetal3D extends the Metal3D framework, which was originally trained on zinc sites, by training on expanded data covering all biologically relevant classes of metal ions.