RetroEA: An efficient evolutionary algorithm for retrosynthetic route planning

Significant progress has been achieved in the field of organic molecular retrosynthesis. However, traditional synthesis methods not only require expert knowledge but are also highly time-consuming. With the development of machine learning, artificial intelligence based methods for organic molecular synthesis have become increasingly popular. Recently, evolutionary optimization emerging as an efficient approach for addressing molecular retrosynthesis, its performance has been hindered by mismatches between the discrete encoding and continuous genetic operator, and the extensive size of the search space. In response to these issues, this paper overcomes the discrete nature of the retrosynthetic route planning problem by employing discrete encoding methods and corresponding discrete genetic operators, and reduces the search space through the use of pruning techniques. Applied to four distinct case products, the proposed method not only reduces the calling of the single-step model by an average of 66.1%, but also decreases the time required to identify three feasible solutions by 51.5% across four cases, outperforming current state-of-the-art methods.