Machine learning differentiates enzymatic and non-enzymatic metals in proteins

Differentiating between enzymatic and non-enzymatic metal-binding sites in proteins is crucial for enzyme identification and design. This study utilizes a large structural dataset of metalloprotein sites and a decision-tree ensemble machine learning model (MAHOMES) to classify metal sites with 92.2% precision and 90.1% recall. The model prioritizes electrostatic and pocket lining features over pocket volume, suggesting the importance of these features in enzymatic activity. MAHOMES outperforms existing methods in differentiating enzymatic from non-enzymatic sequences, promising applications in identifying new enzymatic mechanisms and de novo enzyme design.

Ryan Feehan, Meghan W. Franklin, Joanna S. G. Slusky