MARS: a motif-based autoregressive model for retrosynthesis prediction.

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Discovery

Background:

• Retrosynthesis is crucial for drug discovery, involving the identification of synthetic pathways for target molecules.
• Current graph-generation methods for retrosynthesis face limitations such as high prediction complexity or poor generalization due to their strategies for reactant generation.

Purpose of the Study:

• To propose a novel end-to-end graph generation model for retrosynthesis prediction.
• To address the limitations of existing methods by introducing a more efficient and generalizable approach to reactant generation.

Main Methods:

• Developed a novel end-to-end graph generation model for retrosynthesis.
• The model sequentially identifies reaction centers, generates synthons, and adds chemically meaningful motifs to create reactants.
• Evaluated the model on a benchmark dataset for retrosynthesis prediction.

Main Results:

• The proposed model achieves lower prediction complexity compared to atom-by-atom generation.
• Demonstrates superior performance over methods that add leaving groups, showing better generalization.
• Significantly outperforms previous state-of-the-art models in retrosynthesis prediction.

Conclusions:

• The novel graph generation model is effective for predicting retrosynthesis pathways.
• The approach of adding chemically meaningful motifs enhances prediction accuracy and efficiency.
• The model shows potential as a valuable tool to accelerate drug discovery processes.