Year: 2026
Type: preprint
Abstract: Artificial intelligence (AI) is increasingly recognized as a powerful tool for catalyst optimization in organic synthesis. Yet, applications are often restricted by the need for large datasets, programming expertise, and specialized knowledge in model development. Boronic acidcatalyzed amidation reactions represent an attractive target for AI-guided optimization because of their synthetic relevance, detailed mechanistic understanding, and the need for improved catalysts for amidations. Here, an AI-guided workflow for catalyst optimization in boronic acidcatalyzed amidation reactions was established using the large-language model ChatGPT for code generation and workflow development. A dataset was generated by high-throughput screening of 459 arylboronic acids and combined with DFT-derived steric and electronic descriptors of the catalysts for model training. The model enabled the identification of the unprecedented catalyst FIBA with higher activity than the benchmark catalyst MIBA. In addition, detailed data analysis of the highly active but poorly predicted diboronic acid catalyst DBA led to the identification of a distinct activation mode involving higher-order boronate aggregates that activate and preorganize the amine nucleophile. Thus, improved catalyst performance and new mechanistic insight were obtained through an AI-guided workflow.
Source: ChemRxiv
Authors: Hanna Weller, Tobias Schnitzer
Institution: University of Freiburg
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7220845791