mlDEEPre: Multi-Functional Enzyme Function Prediction With Hierarchical Multi-Label Deep Learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Enzyme Engineering

Background:

• Enzyme function prediction is crucial for enzyme design and disease diagnosis.
• Current methods primarily focus on mono-functional enzymes, neglecting the growing number of multi-functional enzymes.
• Novel computational approaches are needed to address multi-functional enzyme prediction challenges.

Purpose of the Study:

• To introduce mlDEEPre, a novel deep learning method specifically designed for multi-functional enzyme function prediction.
• To improve the accuracy and efficiency of predicting enzyme functionalities, particularly for enzymes with multiple roles.
• To enhance existing deep learning models for enzyme function prediction to accommodate multi-functional enzymes.

Main Methods:

• Developed mlDEEPre, a deep learning model tailored for multi-functional enzyme prediction.
• Implemented a novel loss function that considers label relationships.
• Utilized a self-adapted label assigning threshold for improved prediction accuracy.
• Integrated mlDEEPre with the existing DEEPre model for seamless mono- and multi-functional prediction.

Main Results:

• mlDEEPre demonstrated superior performance in predicting multi-functional enzymes compared to existing methods.
• The method accurately distinguished between mono-functional and multi-functional enzymes.
• mlDEEPre achieved high accuracy in main enzyme class prediction across various criteria.
• The integrated DEEPre model successfully handled both mono- and multi-functional predictions without manual intervention.

Conclusions:

• mlDEEPre offers an accurate and efficient solution for multi-functional enzyme prediction.
• The developed method advances the field of enzyme function prediction, addressing a significant gap in current research.
• The seamless integration capability enhances the utility of deep learning models in enzyme bioinformatics.