MAHOMES II: A webserver for predicting if a metal binding site is enzymatic

Area of Science:

• Computational Biology
• Structural Biology
• Enzymology

Background:

• Protein structure dictates function, but functional annotation is limited for proteins lacking experimental structures.
• Computational methods for functional annotation using predicted structures are needed.
• Distinguishing metalloenzymes from non-metalloenzymes is crucial for understanding enzyme mechanisms.

Purpose of the Study:

• To improve a computational method for identifying metalloenzyme active sites.
• To enable functional annotation of proteins with computationally generated structures.
• To enhance the accuracy and reliability of predicting enzymatic metal sites.

Main Methods:

• Upgraded physicochemical features to reduce reliance on sub-angstrom precision structures.
• Employed machine learning to minimize training data labeling errors.
• Developed and tested a classifier on predicted metalloprotein structures without solved crystal structures or homology.

Main Results:

• The improved classifier achieved 94% precision and 92% recall in identifying enzymatic vs. non-enzymatic metal sites.
• Performance and reliability were enhanced for sites with sub-angstrom variations.
• The model demonstrated 90%-97.5% accuracy on predicted structures, depending on quality.

Conclusions:

• The enhanced method accurately predicts enzymatic metal sites using predicted protein structures.
• Key physicochemical features driving performance include local protein density, residue burial, and pocket accessibility.
• This work facilitates the discovery of novel enzymatic mechanisms and improves de novo metalloenzyme design.