AI-Based Retrosynthesis of Known and Unexplored Covalent Organic Framework Monomers

Abstract

Covalent organic frameworks (COFs), owing to their structural tunability, have a massive design space, of which only a fraction (∼1,300) has been experimentally reported. Using AI-based retrosynthetic tools for the first time in this context, we here address one of the inherent challenges in realizing hypothetical COFs, viz. the synthesis of identified monomers (building blocks). Two general retrosynthetic tools, ASCKOS and AiZynthFinder, showed a combined success rate of 76% in identifying routes to 359 experimentally reported monomers (of 600 2D COFs) with uniform performance across structural motifs. Further, these tools successfully identified synthesis routes to 83 unexplored monomers, capable of forming at least 1,200 hypothetical COFs. This work underlines the efficacy of modern retrosynthesis tools in addressing a challenge in the designed synthesis of COFs.