Template-Free Retrosynthesis with Graph-Prior Augmented Transformers

2025-12-11

AI Summary

Template-free retrosynthetic prediction suffers from low accuracy, poor robustness, and limited practical applicability in synthetic planning. Method: This paper proposes a graph-aware, template-free Transformer framework. It pioneers the direct encoding of molecular graph topology into the self-attention mechanism, enabling joint modeling of SMILES sequences and molecular graphs; introduces graph-enhanced attention and paired-data augmentation to improve generalization; and eliminates reaction templates and rule-based engines, enabling end-to-end generation of interpretable retrosynthetic pathways. Contribution/Results: It is the first work to achieve deep integration of graph-structural priors with the Transformer architecture, substantially enhancing modeling capability for complex reactions. On the USPTO-50K benchmark, it achieves state-of-the-art Top-1 accuracy among template-free methods, significantly outperforming the standard Transformer baseline.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Retrosynthesis reaction prediction seeks to infer plausible reactant molecules for a given product and is a central problem in computer-aided organic synthesis. Despite recent progress, many existing models still fall short of the accuracy and robustness required for practical deployment. This work studies a template-free, Transformer-based framework that eliminates reliance on handcrafted reaction templates or additional chemical rule engines. The model injects molecular graph information into the attention mechanism to jointly exploit SMILES sequences and structural cues, and further applies a paired data augmentation strategy to enhance training diversity and scale. On the USPTO-50K benchmark, our proposed approach achieves state-of-the-art performance among template-free methods and substantially outperforming a vanilla Transformer baseline.

Problem

• Predicts reactant molecules for given products in retrosynthesis.
• Uses graph-augmented Transformers without handcrafted templates.
• Improves accuracy and robustness for practical synthesis applications.

Innovation

• Transformer-based template-free retrosynthesis prediction
• Graph information injected into attention mechanism
• Paired data augmentation enhances training diversity