Accelerating Catalyst Discovery with Artificial Intelligence Using Molecular Dual-Atom Site Catalysts as a Case Study
- 1Institute of Chemistry, University of Tartu, Ravila 14a, Tartu, Estonia, 50411
- 2Department of Chemistry, University of Latvia, Jelgavas iela 1, Riga, Latvia, LV-1004
Abstract
We test whether researchers can delegate the writing of a complete catalyst study to a large language model while retaining scientific judgment. Our case study concerns molecular dual-atom site catalysts for the oxygen reduction reaction, where stabilizing an adsorbed intermediate can make its subsequent reduction less favorable. Using density functional theory calculations on 88 catalysts, we explain this balance through adsorption free energies and the overpotential volcano. We then use adsorbate specific ridge models to predict adsorption energies for 1,260 additional structures and identify candidates for further study. Among candidates with low predicted overpotentials, we find predominantly Fe adsorption centers, with separate structural patterns associated with S bridges and 1,2phenylenediamine. To communicate the reasoning behind these choices, we guided a large language model with explicit scientific writing instructions and refined its explanations through repeated scientific and editorial review. The resulting article is the demonstration: we delegate manuscript preparation while retaining responsibility for its scientific meaning. This approach to accelerating catalyst discovery with artificial intelligence offers researchers an opportunity to concentrate on scientific investigation while guiding how their results are communicated.
Related
- Integrating artificial intelligence and molecular simulation for peptide discovery: limitations, validation and decision-centered workflows
- From Generative Molecular Design to Agentic Artificial Intelligence in Drug Discovery
- Artificial Intelligence‐Assisted Design of High‐Entropy Oxide Catalysts
- Artificial Intelligence-Assisted Dopant Discovery toward Air-Stable Sulfide Solid-State Electrolytes
- Artificial Intelligence in the Discovery and Design of Ionic Liquids for Thermal Energy Applications.
BibTeX
@misc{savchenko2026accelerating,
title = {{Accelerating Catalyst Discovery with Artificial Intelligence Using Molecular Dual-Atom Site Catalysts as a Case Study}},
author = {Denis Savchenko and Bruno Ina cio da Silva Roux Leite and Timmo-Henrik Pukk and Ba Long Nguyen and Vladislav Ivanistsev and Nadezda Kongi},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010768/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010768/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Accelerating Catalyst Discovery with Artificial Intelligence Using Molecular Dual-Atom Site Catalysts as a Case Study
AU - Denis Savchenko
AU - Bruno Ina cio da Silva Roux Leite
AU - Timmo-Henrik Pukk
AU - Ba Long Nguyen
AU - Vladislav Ivanistsev
AU - Nadezda Kongi
PY - 2026
DA - 2026/10/21
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010768/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010768/v1
AB - We test whether researchers can delegate the writing of a complete catalyst study to a large language model while retaining scientific judgment. Our case study concerns molecular dual-atom site catalysts for the oxygen reduction reaction, where stabilizing an adsorbed intermediate can make its subsequent reduction less favorable. Using density functional theory calculations on 88 catalysts, we explain this balance through adsorption free energies and the overpotential volcano. We then use adsorbate specific ridge models to predict adsorption energies for 1,260 additional structures and identify candidates for further study. Among candidates with low predicted overpotentials, we find predominantly Fe adsorption centers, with separate structural patterns associated with S bridges and 1,2phenylenediamine. To communicate the reasoning behind these choices, we guided a large language model with explicit scientific writing instructions and refined its explanations through repeated scientific and editorial review. The resulting article is the demonstration: we delegate manuscript preparation while retaining responsibility for its scientific meaning. This approach to accelerating catalyst discovery with artificial intelligence offers researchers an opportunity to concentrate on scientific investigation while guiding how their results are communicated.
ER -