RPBP: Deep Retrosynthesis Reaction Prediction Based on Byproducts

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Retrosynthesis prediction is vital for designing synthetic routes in organic synthesis and drug discovery.
• Data-driven deep learning models have advanced retrosynthesis prediction, but often overlook byproduct information.
• Existing benchmark datasets lack comprehensive byproduct data, limiting model performance.

Purpose of the Study:

• To develop a novel two-stage retrosynthesis prediction framework that explicitly incorporates byproduct information.
• To enhance the accuracy and efficiency of predicting synthetic routes by leveraging byproduct data.
• To demonstrate the practical utility of the proposed framework in accelerating drug discovery.

Main Methods:

• Proposed a two-stage retrosynthesis reaction prediction framework named RPBP.
• RPBP first predicts the byproduct of a reaction based on the product molecule.
• Subsequently, it predicts reactants using both the product and the predicted byproduct, considering reagents, conditions, and reaction sites.

Main Results:

• The RPBP model achieved 54.7% top-1 accuracy for unknown reaction classes and 66.6% for known reaction classes.
• RPBP demonstrated superior performance compared to existing methods, particularly for known-class reactions.
• The framework showed reduced model learning complexity in natural language processing (NLP) tasks by incorporating byproduct information.

Conclusions:

• Incorporating byproduct information significantly enhances retrosynthesis prediction accuracy.
• The RPBP framework offers a practical and effective approach for accelerating drug discovery and organic synthesis.
• The model's ability to predict kinase drugs from literature highlights its potential for real-world applications.