Predicting enzyme-compound associations for enzyme-catalysed reactions.

Area of Science:

• Biochemistry and Cheminformatics
• Computational Biology
• Environmental Science

Background:

• Enzyme-catalyzed reactions are crucial in pharmaceutical metabolism and biodegradation.
• Predicting reaction outcomes is vital for identifying toxic by-products and regulatory approval.
• Current prediction methods often require known enzyme information, which is frequently unavailable in complex environments like biodegradation.

Purpose of the Study:

• To develop a novel framework for predicting enzyme-compound associations without prior enzyme knowledge.
• To evaluate the performance of a hierarchical multi-label classifier for this prediction task.
• To investigate the impact of enzyme information on the accuracy of chemical product prediction.

Main Methods:

• Developed and evaluated a hierarchical multi-label classifier to predict enzyme commission number-compound associations.
• Implemented a self-tuning mechanism for optimal hyperparameter selection.
• Compared product prediction performance with and without the inclusion of enzyme data.

Main Results:

• Achieved a hierarchical F1-score of up to 93.2%, outperforming existing methodologies.
• Demonstrated that including enzyme commission numbers improves product prediction performance by approximately two percentage points in biodegradation case studies.
• The proposed method shows superior performance compared to existing approaches.

Conclusions:

• The novel hierarchical multi-label classifier framework effectively predicts enzyme-compound associations, even when specific enzymes are unknown.
• Incorporating enzyme information, whether true or predicted, enhances the accuracy of chemical product prediction.
• This work offers a significant advancement in predicting enzymatic reactions for applications in biodegradation and pharmaceutical metabolism.