.png){kind=icon}
.png){kind=icon}
Return
Meta-learning accelerates multi-objective Bayesian optimization of chemical reaction: a monoacylation case study
.png){kind=icon}
PRE
AI
G
J
R
Y
W
A
DOI：10.1016/j.jiec.2026.03.030.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The monoacylated diamine moiety is a common structural fragment in pharmaceuticals, but its synthesis often requires complex, multi-step procedures. Although multi-objective Bayesian optimization (MOBO) is a powerful method for optimizing chemical reactions involving competing objectives, it rarely leverages data from related reactions to accelerate the optimization process. In this work, we present Meta Learning-assisted Multi-objective Bayesian Optimization (Meta-MOBO), a novel framework that integrates meta-learning with MOBO to effectively transfer knowledge from historical experimental data to new multi-objective chemical reaction optimization tasks. We first benchmarked Meta-MOBO across four computational case studies, reducing the number of required iterations by 25–45% compared to conventional MOBO while achieving an average improvement of 2.92% in hypervolume. Meta-MOBO was then applied to optimize yield and productivity in monoacylation reactions across five distinct substrates using a continuous flow reactor. This approach successfully identified Pareto frontiers for novel substrates while significantly reducing the number of exploratory experiments typically required in standard MOBO workflows. These findings establish Meta-MOBO as an efficient and cost-effective strategy for multi-objective chemical reaction optimization.
Keywords:
Meta-MOBO
Multi-objective Bayesian optimization
Meta-learning
Chemical reaction optimization
Monoacylation
Journal
IF:
6
Papers:
8.6K
Citations:
3.2W
Organization
Cited Papers