Retrosynthesis Prediction via Search in (Hyper) Graph

• DOI: 10.48550/arxiv.2402.06772
• Publication Date: 2024-02-09
• Authors: Zixun Lan; Binjie Hong; Jiajun Zhu; Zuo Zeng; Zhenfu Liu; Limin Yu; Fei Ma
• Keywords: Retrosynthetic analysis

Predicting reactants from a specified core product stands as fundamental challenge within organic synthesis, termed retrosynthesis prediction. Recently, semi-template-based methods and graph-edits-based have achieved good performance in terms of both interpretability accuracy. However, due to their mechanisms these cannot predict complex reactions, e.g., reactions with multiple reaction center or attaching the same leaving group more than one atom. In this study we propose method, \textbf{Retro}synthesis via \textbf{S}earch \textbf{i}n (Hyper) \textbf{G}raph (RetroSiG) framework alleviate limitations. proposed turn identification completion tasks searching molecular graph hypergraph respectively. As method RetroSiG has several advantages. First, is able handle mentioned above by its novel search mechanism. Second, naturally exploits model implicit dependencies between groups. Third, makes full use prior, i.e., one-hop constraint. It reduces space enhances overall performance. Comprehensive experiments demonstrated that competitive results. Furthermore, conducted show capability predicting reactions. Ablation verified efficacy specific elements, such constraint hypergraph.