Retrosynthesis prediction using an end-to-end graph generative architecture for molecular graph editing.

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Artificial Intelligence

Background:

• Retrosynthesis planning is crucial for organic synthesis but remains challenging.
• Current deep learning methods for retrosynthesis prediction have limitations in applicability and interpretability.
• Improving predictive accuracy is essential for practical applications.

Purpose of the Study:

• To develop an end-to-end deep learning architecture for retrosynthesis prediction.
• To enhance the applicability and interpretability of retrosynthesis prediction models.
• To achieve state-of-the-art performance in semi-template-based retrosynthesis.

Main Methods:

• Developed Graph2Edits, an end-to-end architecture based on graph neural networks.
• Employed an auto-regressive approach to predict graph edits for retrosynthesis.
• Integrated two-stage processes into a one-pot learning strategy for improved efficiency.

Main Results:

• Graph2Edits demonstrated improved applicability, especially for complex reactions.
• The model provides more interpretable predictions compared to existing methods.
• Achieved a state-of-the-art top-1 accuracy of 55.1% on the USPTO-50k dataset.

Conclusions:

• Graph2Edits offers a promising advancement in computer-aided synthesis planning.
• The arrow-pushing formalism-inspired approach enhances retrosynthesis prediction.
• This method represents a significant step towards practical and interpretable AI in organic synthesis.