Year: 2026
Type: preprint
Abstract: Modern computer-aided synthesis planning (CASP) tools, particularly retrosynthesis planning, are becoming practical aids for the ideation stage of chemical synthesis design. However, the same target molecule can be synthesized via substantially different routes depending on the purpose: a discovery or medicinal chemistry route used to obtain small quantities during early development, or a manufacturing route optimized for industrial-scale production. While CASP tools have been extensively studied and benchmarked for the former, their ability to support manufacturing route design remains largely unexplored. Performance of CASP tools is usually measured by benchmark experiments, which typically report the fraction of target molecules for which a tool can generate a synthesis route, connecting the target molecule through a sequence of intermediates to starting materials. Sometimes, to assess the quality of the generated routes, they are also manually compared with expert-designed routes reported in the literature. Currently, public benchmark datasets focused on process and manufacturing chemistry are lacking. In this study, we take a first step in this direction by analyzing one of the main sources of process chemistry research, the Organic Process Research & Development (OPRD) journal. We reviewed OPRD papers published between 1997 and 2022 and selected 2507 papers reporting multi-step synthesis routes. We then separately collected and digitized the target molecules of corresponding synthesis routes as SMILES strings. This resulted in a collection of 2219 unique target molecules, each paired with the DOI of the original publication reporting their corresponding synthesis route. This benchmark can therefore be used to assess how many process-chemistry targets modern CASP tools can solve and, for solved targets, to compare predicted routes with the expert-designed routes reported in the literature. The dataset is publicly available at: https://github.com/KagakuLab/OPRDbench.
Source: ChemRxiv
Authors: D. Zankov, M. Barbatti
Language: English
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7221797331