G²Retro as a two-step graph generative models for retrosynthesis prediction.

Area of Science:

• Computational Chemistry
• Organic Synthesis
• Machine Learning

Background:

• Retrosynthesis is crucial for identifying chemical synthesis routes by transforming target molecules into reactants.
• Computational methods are increasingly used to expedite the design of retrosynthetic pathways.
• Existing methods face challenges in accurately predicting precursors for complex molecules.

Purpose of the Study:

• To develop a novel generative framework, G2Retro, for accurate one-step retrosynthesis prediction.
• To improve the efficiency and accuracy of identifying reactants from target products in chemical synthesis.
• To establish a new benchmark for computational retrosynthesis prediction.

Main Methods:

• G2Retro employs a generative approach that mimics the reversed logic of synthetic reactions.
• It predicts reaction centers within target molecules (products) and identifies necessary synthons.
• The framework learns from molecular graphs to predict reaction centers and completes synthons into reactants by considering product structures and optimal attachment paths.

Main Results:

• G2Retro demonstrates superior performance in predicting reactants for given products compared to state-of-the-art methods on a benchmark dataset.
• The framework successfully identifies reaction centers and generates plausible synthons and reactants.
• The approach shows robustness in handling diverse molecular structures.

Conclusions:

• G2Retro offers a significant advancement in computational retrosynthesis, improving prediction accuracy.
• The generative framework provides a powerful tool for accelerating the design of chemical synthesis routes.
• This work paves the way for more efficient drug discovery and chemical development.