From Retrosynthetic Prediction to the Bench: Experimental Evaluation of Computer-Generated Synthetic Routes
- 1Department of Chemistry and Bioscience, Aalborg University, Fredrik Bajers Vej 7H, Aalborg, Denmark, DK-9220
- 2Department of Physics, Chemistry and Bioscience, University of Southern Denmark, Campusvej 55, Odense, Denmark, DK-5230
Abstract
Retrosynthesis tools propose bond disconnections and precursor assignments, but the chemist must still evaluate these choices and select reaction conditions and purification before a route can be run. In this work, we examine 53 PostEra Manifold and AiZynthFinder proposals for three small organic targets by combining chemist assessments with selected synthesis outcomes. Proposals were rejected or deprioritised for different reasons: incorrect connectivity, unsupported transformations, or extra steps and costly precursors in routes that would otherwise work. For a methoxy-substituted biaryl ketone, a Suzuki–Miyaura coupling using the chemist-preferred assignment of coupling partners gave 94% isolated yield. For a naphthylphenol, coupling with 2hydroxyphenylboronic acid gave 92% isolated yield with an aryl bromide and 7% with the corresponding chloride under the same procedure. The chloride result supports the chemist’s preference for the bromide, whereas the 92% yield shows that the earlier concern about this boronic acid, based on its behaviour in water, was not limiting under the tested conditions. We therefore propose that each generated route be judged on four separate points: whether it gives the correct target, whether the transformation is chemically supported, whether the route is practical, and whether it has been tested experimentally.
Related
- Experimental and numerical study of shock wave propagation in water generated by pulsed arc electrohydraulic discharges
- Application of spectroscopic and multispectral imaging technologies on the assessment of ready-to-eat pineapple quality: A performance evaluation study of machine learning models generated from two commercial data analytics tools
- Near-Complete Synthon Library via Retrosynthetic Analysis for Automated Organic Synthesis
- Rachel: A general-purpose language model directs and revises retrosynthetic routes
- Leveraging the Condensed Graph of Reaction for Clustering Retrosynthetic Pathways
BibTeX
@misc{dandrea2026from,
title = {{From Retrosynthetic Prediction to the Bench: Experimental Evaluation of Computer-Generated Synthetic Routes}},
author = {Laura d’Andrea and Heidi Fuglsang Christiansen and Casper Steinmann},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010071/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010071/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - From Retrosynthetic Prediction to the Bench: Experimental Evaluation of Computer-Generated Synthetic Routes
AU - Laura d’Andrea
AU - Heidi Fuglsang Christiansen
AU - Casper Steinmann
PY - 2026
DA - 2026/10/07
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010071/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010071/v1
AB - Retrosynthesis tools propose bond disconnections and precursor assignments, but the chemist must still evaluate these choices and select reaction conditions and purification before a route can be run. In this work, we examine 53 PostEra Manifold and AiZynthFinder proposals for three small organic targets by combining chemist assessments with selected synthesis outcomes. Proposals were rejected or deprioritised for different reasons: incorrect connectivity, unsupported transformations, or extra steps and costly precursors in routes that would otherwise work. For a methoxy-substituted biaryl ketone, a Suzuki–Miyaura coupling using the chemist-preferred assignment of coupling partners gave 94% isolated yield. For a naphthylphenol, coupling with 2hydroxyphenylboronic acid gave 92% isolated yield with an aryl bromide and 7% with the corresponding chloride under the same procedure. The chloride result supports the chemist’s preference for the bromide, whereas the 92% yield shows that the earlier concern about this boronic acid, based on its behaviour in water, was not limiting under the tested conditions. We therefore propose that each generated route be judged on four separate points: whether it gives the correct target, whether the transformation is chemically supported, whether the route is practical, and whether it has been tested experimentally.
ER -