Geometric deep learning-enabled metal-binding site identification and grafting

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Metal-binding sites are crucial for protein function, including catalysis, stability, and interactions.
• Understanding these sites is key to discovering novel protein functions and engineering new metalloproteins.

Purpose of the Study:

• To introduce MeSiteIG, a deep learning tool for identifying and grafting metal-binding sites.
• To leverage geometric deep learning for predicting metal-binding residues based on geometric conservation.

Main Methods:

• MeSiteIG utilizes three modules: residue triplet searcher (ResTriS), metal-binding site identifier (MeSI), and metal-binding site alignment (MeSA).
• The core MeSI module employs E3-equivarient graph neural networks for residue prediction, independent of coordinating residue types and side-chain details.
• The tool operates at a speed of approximately 300 samples per second.

Main Results:

• MeSI demonstrated superior performance on an independent test set.
• MeSiteIG successfully identified previously overlooked protein metal-binding sites and potential bimetallic sites.
• The tool enabled the grafting of metal sites onto antibody surfaces and protein pockets.

Conclusions:

• MeSiteIG is an efficient and effective tool for metal-binding site identification and engineering.
• The ability to graft metal sites opens avenues for designing novel metalloproteins with tailored functions.
• This work advances the field of metalloprotein design and discovery.