Accelerating discovery of phosphorescent blue OLED emitters using chemical foundation models
13 affiliations
- 1Department of Physical and Environmental Sciences
- 2University of Toronto Scarborough
- 3Alliance for AI-Accelerated Materials Discovery
- 4Mila–Quebec AI Institute
- 5Department of Electrical and Computer Engineering
- 6University of Toronto
- 7Organic Semiconductor Centre, EaStCHEM School of Chemistry
- 8University of St Andrews
- 9Johns Hopkins University Scientific Software Engineering Center
- 10McCormick School of Engineering and Applied Science
- 11Northwestern University
- 12Paula M. Trienens Institute for Sustainability and Energy
- 13Department of Chemistry
Abstract
In OLED displays, phosphorescent emitter molecules are used for red and green but not for blue pixels. Iridium organometallics are promising phosphorescent emitters, but a molecule satisfying the requirements of color, brightness, and stability has not yet been developed. One strategy to discover an emitter is virtual screening, in which computational predictions are used to prioritize candidates for synthesis. Due to the computational difficulty of screening with first-principles methods such as density functional theory (DFT), empirical machine-learning methods have been used for virtual screening. Recently, machine-learned interatomic potentials (MLIPs) have provided fast proxies for DFT energies. This suggests that virtual screening could be conducted with ML proxies for DFT calculations rather than empirical ML, but this has not yet been applied to OLEDs. We show on a dataset of experimental values extracted from the literature that for emission wavelength, MLIP-based predictions are not inferior to empirical ML. Using an MLIP, we screened millions of molecules for blue emission. Although blue emission is not sufficient for an emitter, our results suggest that further development of virtual screening could focus on machine learned proxies for DFT prediction pipelines rather than models trained to experimental data.
Related
- Accelerating Mechanistic Understanding and Prediction of Chemical Selectivity with Foundation Machine-Learned Interatomic Potentials
- Accelerating the discovery of nanoporous gas separation materials via Coulomb-enhanced classical density functional theory
- Computationally Accelerated Discovery of Efficient Dual Component White Light Emitters
- Accelerating Catalyst Discovery with Artificial Intelligence Using Molecular Dual-Atom Site Catalysts as a Case Study
- Bgolearn: a unified Bayesian optimization framework for accelerating materials discovery
BibTeX
@misc{davis2026accelerating,
title = {{Accelerating discovery of phosphorescent blue OLED emitters using chemical foundation models}},
author = {Alexander Davis and Ihor Neporozhnii and Félix Therrien and Suhas Mahesh and Zhibo Wang and Ze-Lin Zhu and Jeffrey Cheung and Oleksandra Ostapenko and Alex Zhuang and Xiang Chen and Arik Mitschang and Edward Hunter and Brandon Sutherland and Sjoerd Hoogland and Edward Sargent and Eli Zysman-Colman and Oleksandr Voznyy},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15003869/v2},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15003869/v2},
note = {CC BY},
}
RIS
TY - UNPB
TI - Accelerating discovery of phosphorescent blue OLED emitters using chemical foundation models
AU - Alexander Davis
AU - Ihor Neporozhnii
AU - Félix Therrien
AU - Suhas Mahesh
AU - Zhibo Wang
AU - Ze-Lin Zhu
AU - Jeffrey Cheung
AU - Oleksandra Ostapenko
AU - Alex Zhuang
AU - Xiang Chen
AU - Arik Mitschang
AU - Edward Hunter
AU - Brandon Sutherland
AU - Sjoerd Hoogland
AU - Edward Sargent
AU - Eli Zysman-Colman
AU - Oleksandr Voznyy
PY - 2026
DA - 2026/10/07
PB - ChemRxiv
DO - 10.26434/chemrxiv.15003869/v2
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15003869/v2
AB - In OLED displays, phosphorescent emitter molecules are used for red and green but not for blue pixels. Iridium organometallics are promising phosphorescent emitters, but a molecule satisfying the requirements of color, brightness, and stability has not yet been developed. One strategy to discover an emitter is virtual screening, in which computational predictions are used to prioritize candidates for synthesis. Due to the computational difficulty of screening with first-principles methods such as density functional theory (DFT), empirical machine-learning methods have been used for virtual screening. Recently, machine-learned interatomic potentials (MLIPs) have provided fast proxies for DFT energies. This suggests that virtual screening could be conducted with ML proxies for DFT calculations rather than empirical ML, but this has not yet been applied to OLEDs. We show on a dataset of experimental values extracted from the literature that for emission wavelength, MLIP-based predictions are not inferior to empirical ML. Using an MLIP, we screened millions of molecules for blue emission. Although blue emission is not sufficient for an emitter, our results suggest that further development of virtual screening could focus on machine learned proxies for DFT prediction pipelines rather than models trained to experimental data.
ER -