Machine learning differentiates enzymatic and non-enzymatic metals in proteins

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• Metalloenzymes constitute 40% of all known enzymes and catalyze diverse reactions.
• Distinguishing active metalloenzymes from inactive metal-binding sites is challenging due to similar physicochemical properties.
• Identifying these differences is critical for discovering new enzymes and designing novel ones.

Purpose of the Study:

• To develop a computational model for classifying metalloprotein sites as enzymatic or non-enzymatic.
• To identify key physicochemical features differentiating catalytic and non-catalytic metal-binding sites.
• To enhance the accuracy of enzyme identification and de novo enzyme design.

Main Methods:

• Compiled the largest structural dataset of enzymatic and non-enzymatic metalloprotein sites.
• Utilized a decision-tree ensemble machine learning model for classification.
• Evaluated model performance using precision and recall metrics.

Main Results:

• Achieved 92.2% precision and 90.1% recall in classifying metalloprotein sites.
• Electrostatic and pocket lining features were identified as more important than pocket volume.
• The developed model outperformed existing methods in differentiating enzymatic from non-enzymatic sequences.

Conclusions:

• The machine learning model accurately distinguishes enzymatic from non-enzymatic metalloprotein sites.
• Key features for enzymatic activity prediction include electrostatic and pocket lining characteristics.
• This approach facilitates the identification of novel enzymatic mechanisms and supports de novo enzyme design.