Graph attention framework for site-selectivity prediction in undirected C(sp3)–H borylation
Year: 2026
Type: preprint
Abstract: C–H borylation has surfaced as a powerful transformation in molecular syntheses, providing direct access to organoboron compounds – highly applicable to pharmaceutical and crop protection industries, as well as materials science. While site-selectivity prediction is established for C(sp 2 )–H borylation, significantly more challenging C(sp 3 )–H borylations remain largely unexplored. In sharp contrast, we have developed a graph-attention deep-learning framework combining molecular representations with quantum-mechanical descriptors to predict relative C(sp 3 )–H borylation site selectivities. Using curated literature reactions and a site labeling algorithm, we evaluated various substrate representations and reaction-condition encodings. Bipartite encoding of reaction conditions outperformed independent component encodings. Validation on previously unreported substrates reproduced experimental selectivity distributions, supporting the framework’s robustness and practical utility.
Source: ChemRxiv
Institution: University of Göttingen
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
Award: G1633987393
OpenAlex ID: W7211928545