Abstract:Metals play vital roles in both the mechanism and architecture of biological macromolecules, and are the most frequently encountered ligands (i.e. non-solvent heterogeneous chemical atoms) in the determination of macromolecular crystal structures. However, metal coordinating environments in protein structures are not always easy to check in routine validation procedures, resulting in an abundance of misidentified and/or suboptimally modeled metal ions in the Protein Data Bank (PDB). We present a solution to identify these problems in three distinct yet related aspects: (1) coordination chemistry; (2) agreement of experimental B-factors and occupancy; and (3) the composition and motif of the metal binding environment. Due to additional strain introduced by macromolecular backbones, the patterns of coordination of metal binding sites in metal-containing macromolecules are more complex and diverse than those found in inorganic or organometallic chemistry. These complications make a comprehensive library of "permitted" coordination chemistry in protein structures less feasible, and the usage of global parameters such as the bond valence method more practical, in the determination and validation of metal binding environments. Although they are relatively infrequent, there are also cases where the experimental B-factor or occupancy of a metal ion suggests careful examination. We have developed a web-based tool called CheckMyMetal [1](http://csgid.org/csgid/metal_sites/) for the quick validation of metal binding sites. Moreover, the acquired knowledge of the composition and spatial arrangement (motif) of the coordinating atoms around the metal ion may also help in the modeling of metal binding sites in macromolecular structures. All of the studies described herein were performed using the NEIGHBORHOOD SQL database [2], which connects information about all modeled non-solvent heterogeneous chemical motifs in PDB structure by vectors describing all contacts to neighboring residues and atoms. NEIGHBORHOOD has broad applications for the validation and data mining of ligand binding environments in the PDB.

Predicting metal-protein interactions using cofolding methods: Status quo

Metal3D: a general deep learning framework for accurate metal ion location prediction in proteins

Prediction of the Iron–Sulfur Binding Sites in Proteins Using the Highly Accurate Three-Dimensional Models Calculated by AlphaFold and RoseTTAFold

Challenges in predicting PROTAC-mediated Protein-Protein Interfaces with AlphaFold

MetalExplorer, a Bioinformatics Tool for the Improved Prediction of Eight Types of Metal-Binding Sites Using a Random Forest Algorithm with Two- Step Feature Selection

Inferring metal binding sites in flexible regions of proteins

How accurately can one predict drug binding modes using AlphaFold models?

Prediction of structures of zinc-binding proteins through explicit modeling of metal coordination geometry.

Do Deep Learning Models for Co-Folding Learn the Physics of Protein-Ligand Interactions?

Predicting metal-binding proteins and structures through integration of evolutionary-scale and physics-based modeling

Improved prediction of protein-protein interactions using AlphaFold2

A Perspective on the Prospective Use of AI in Protein Structure Prediction

Check Your Metal - Not Every Density Blob is A Water Molecule

PinMyMetal: A hybrid learning system to accurately model metal binding sites in macromolecules.

ColabDock: Inverting AlphaFold Structure Prediction Model for Protein-Protein Docking with Experimental Restraints

Assessment of AI-Based Protein Structure Prediction for the NLRP3 Target

Predicting the Structure of Enzymes with Metal Cofactors:The Example of [FeFe] Hydrogenases

A Novel Prediction Method for Metal-Ion Binding Sites in Protein Sequence Based on Ensemble Learning

Reliable protein-protein docking with AlphaFold, Rosetta and replica-exchange

AI-Based Protein Structure Prediction in Drug Discovery: Impacts and Challenges

Data Mining of Metal Ion Environments Present in Protein Structures