.png){kind=icon}
Return
MBERA: Multi-Branch Evolutionary Retrosynthesis Algorithm
.png){kind=icon}
PRE
AI
DOI：10.1016/j.swevo.2026.102374.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multi-step retrosynthetic planning remains challenging due to the inherent multi-branch structure of real chemical syntheses, where multiple precursor sets may be coordinated. Existing search algorithms operate primarily on single root-to-leaf trajectories, making them ill-suited for reactions requiring parallel decomposition. To address this limitation, we propose a multi-branch evolutionary algorithm, a discrete evolutionary formulation designed specifically for multi-precursor retrosynthesis. It introduces a multi-branch genome encoding that captures different branch reactant expansions, enabling efficient navigation of combinatorial route spaces. A branch genetic operator further supports coordinated crossover and mutation across precursor paths. Across representative molecules, multi-branch evolutionary algorithm reduces single-step model calls by 70%–95% compared with classical tree-search algorithms and by more than 98% on the most challenging cases. The method exhibits stable behavior across different solution requirements and delivers improvements over established retrosynthetic search techniques.
Keywords:
Multi-step retrosynthesis
Multi-branch genome encoding
Evolutionary algorithm
Retrosynthetic planning
Combinatorial route space
Journal
IF:
8.5
Papers:
2.2K
Citations:
1.0W
Organization
Cited Papers
ShareSaveTowards efficient discovery of green synthetic pathways with Monte Carlo tree search and reinforcement learning.png)
CHEMICAL SCIENCE
OAAI
CHEMICAL SCIENCEIF7.4
OAAIShareSaveRetrosynthesis prediction with an interpretable deep-learning framework based on molecular assembly tasksNATURE COMMUNICATIONSOAAI
NATURE COMMUNICATIONSIF15.7
OAAIShareSaveShareSave
View more