Design and experimental validation of a photocatalyst recommender based on a large language model

Francis Millward; Michal Kulczykowski; Jay Badland-Shaw; Sara Szymkuc; Rajan Suraksha; Aniket Kumar Srivastawa; Violaine Manet; Máire Griffin; Megan Bryden; Thomas Comerford; Lea Hämmerling; Aminata Mariko; Bartosz A. Grzybowski; Eli Zysman-Colman

doi:10.1002/anie.202514544

Design and experimental validation of a photocatalyst recommender based on a large language model

Francis Millward, Michal Kulczykowski, Jay Badland-Shaw, Sara Szymkuc, Rajan Suraksha, Aniket Kumar Srivastawa, Violaine Manet, Máire Griffin, Megan Bryden, Thomas Comerford, Lea Hämmerling, Aminata Mariko, Bartosz A. Grzybowski^*, Eli Zysman-Colman^*

^*Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

Abstract

Utilizing an extensive library of literature on photocatalytic transformations, we disclose the development of a machine learning (ML) model for the recommendation of photocatalysts most suitable for reactions of interest. The model is trained on > 36 000 such literature examples and uses an architecture inspired by the Bidirectional Encoder Representations from Transformer (BERT) large language model. Under cross-validation, it can suggest the “correct” photocatalysts with ∼90% accuracy. When experimentally tested on five out-of-box reactions, this algorithm consistently suggested photocatalysts that gave yields competitive to those chosen by human researchers and frequently suggested alternative photocatalysts that are potentially more appealing than the originally selected photocatalyst. Altogether, this platform serves as a valuable tool for researchers undertaking reaction optimization programs. The model is free to use at https://photocatals.grzybowskigroup.pl/predict/.

Original language	English
Article number	e14544
Number of pages	10
Journal	Angewandte Chemie International Edition
Volume	Early View
Early online date	9 Dec 2025
DOIs	https://doi.org/10.1002/anie.202514544
Publication status	E-pub ahead of print - 9 Dec 2025

Keywords

Large language models
Machine learning
Photocatalysis

Access to Document

10.1002/anie.202514544Licence: CC BY

Millward_2025_ACIE_Design-photocatalyst-recommender-LLM_CC
© 2025 The Author(s). This is an open access article under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits use, distribution and reproduction in any medium, provided the original work is properly cited.
Final published version, 1.97 MBLicence: CC BY

2 Finished
3 Active

SolOLED: TADF Dendrimers for Highly Efficient Solution-Processed OLEDs
Zysman-Colman, E. (PI)
EPSRC
1/01/25 → 31/12/29
Project: Fellowship
MechanophotocatalysiS: Mechanophotocatalysis: a new frontier for solvent-minimized photocatalysis react
Zysman-Colman, E. (PI)
The Leverhulme Trust
1/09/23 → 31/08/27
Project: Standard
Allan Watson Programme Grant: Boron: Beyond the Reagent
Watson, A. (PI), Morris, R. (CoI), Smith, A. (CoI) & Zysman-Colman, E. (CoI)
UK Research and Innovation
1/05/23 → 30/04/28
Project: Standard

Design and Experimental Validation of a Photocatalyst Recommender based on a Large Language Model (dataset)
Millward, F. D. (Creator), Kulczykowski, M. (Creator), Badland-Shaw, J. (Creator), Szymkuc, S. (Creator), Suraksha, R. (Creator), Kumar Srivastawa, A. (Creator), Manet, V. (Creator), Griffin, M. A. (Creator), Bryden, M. A. (Creator), Comerford, T. (Creator), Hammerling, L. (Creator), Mariko, A. (Creator), Grzybowski, B. A. (Creator) & Zysman-Colman, E. (Creator), University of St Andrews, 16 Dec 2025
DOI: 10.17630/21981110-a42c-44cd-91aa-b5c52a60a7b7
Dataset

File

Cite this

Millward, F., Kulczykowski, M., Badland-Shaw, J., Szymkuc, S., Suraksha, R., Kumar Srivastawa, A., Manet, V., Griffin, M., Bryden, M., Comerford, T., Hämmerling, L., Mariko, A., Grzybowski, B. A., & Zysman-Colman, E. (2025). Design and experimental validation of a photocatalyst recommender based on a large language model. Angewandte Chemie International Edition, Early View, Article e14544. Advance online publication. https://doi.org/10.1002/anie.202514544

@article{946d66a3a33d44c782b224e7d550a8b0,

title = "Design and experimental validation of a photocatalyst recommender based on a large language model",

abstract = "Utilizing an extensive library of literature on photocatalytic transformations, we disclose the development of a machine learning (ML) model for the recommendation of photocatalysts most suitable for reactions of interest. The model is trained on > 36 000 such literature examples and uses an architecture inspired by the Bidirectional Encoder Representations from Transformer (BERT) large language model. Under cross-validation, it can suggest the “correct” photocatalysts with ∼90\% accuracy. When experimentally tested on five out-of-box reactions, this algorithm consistently suggested photocatalysts that gave yields competitive to those chosen by human researchers and frequently suggested alternative photocatalysts that are potentially more appealing than the originally selected photocatalyst. Altogether, this platform serves as a valuable tool for researchers undertaking reaction optimization programs. The model is free to use at https://photocatals.grzybowskigroup.pl/predict/.",

keywords = "Large language models, Machine learning, Photocatalysis",

author = "Francis Millward and Michal Kulczykowski and Jay Badland-Shaw and Sara Szymkuc and Rajan Suraksha and \{Kumar Srivastawa\}, Aniket and Violaine Manet and M{\'a}ire Griffin and Megan Bryden and Thomas Comerford and Lea H{\"a}mmerling and Aminata Mariko and Grzybowski, \{Bartosz A.\} and Eli Zysman-Colman",

note = "Funding: M.K. and S.S. were supported by Allchemy, Inc. Analysis of results and writing of the paper by B.A.G. was supported by the Institute for Basic Science, Korea (project code IBS-R020-D1). The UK authors thank the Leverhulme Trust for support (RPG-2023–110), the Engineering and Physical Sciences Research Council for funding (EP/W007517/1, EP/W015137/1, EP/Z535291/1, and EP/W522259/1), the European Commission (PhotoReAct ITN: 956324), Syngenta and Johnson Matthey for support. F.M. M.A.B., and T.C. thank the EaSI-CAT CDT at the University of St Andrews for support in the form of a studentship.",

year = "2025",

month = dec,

day = "9",

doi = "10.1002/anie.202514544",

language = "English",

volume = "Early View",

journal = "Angewandte Chemie International Edition",

issn = "1433-7851",

publisher = "John Wiley \& Sons, Ltd",

}

Millward, F, Kulczykowski, M, Badland-Shaw, J, Szymkuc, S, Suraksha, R, Kumar Srivastawa, A, Manet, V, Griffin, M, Bryden, M, Comerford, T, Hämmerling, L, Mariko, A, Grzybowski, BA & Zysman-Colman, E 2025, 'Design and experimental validation of a photocatalyst recommender based on a large language model', Angewandte Chemie International Edition, vol. Early View, e14544. https://doi.org/10.1002/anie.202514544

TY - JOUR

T1 - Design and experimental validation of a photocatalyst recommender based on a large language model

AU - Millward, Francis

AU - Kulczykowski, Michal

AU - Badland-Shaw, Jay

AU - Szymkuc, Sara

AU - Suraksha, Rajan

AU - Kumar Srivastawa, Aniket

AU - Manet, Violaine

AU - Griffin, Máire

AU - Bryden, Megan

AU - Comerford, Thomas

AU - Hämmerling, Lea

AU - Mariko, Aminata

AU - Grzybowski, Bartosz A.

AU - Zysman-Colman, Eli

N1 - Funding: M.K. and S.S. were supported by Allchemy, Inc. Analysis of results and writing of the paper by B.A.G. was supported by the Institute for Basic Science, Korea (project code IBS-R020-D1). The UK authors thank the Leverhulme Trust for support (RPG-2023–110), the Engineering and Physical Sciences Research Council for funding (EP/W007517/1, EP/W015137/1, EP/Z535291/1, and EP/W522259/1), the European Commission (PhotoReAct ITN: 956324), Syngenta and Johnson Matthey for support. F.M. M.A.B., and T.C. thank the EaSI-CAT CDT at the University of St Andrews for support in the form of a studentship.

PY - 2025/12/9

Y1 - 2025/12/9

N2 - Utilizing an extensive library of literature on photocatalytic transformations, we disclose the development of a machine learning (ML) model for the recommendation of photocatalysts most suitable for reactions of interest. The model is trained on > 36 000 such literature examples and uses an architecture inspired by the Bidirectional Encoder Representations from Transformer (BERT) large language model. Under cross-validation, it can suggest the “correct” photocatalysts with ∼90% accuracy. When experimentally tested on five out-of-box reactions, this algorithm consistently suggested photocatalysts that gave yields competitive to those chosen by human researchers and frequently suggested alternative photocatalysts that are potentially more appealing than the originally selected photocatalyst. Altogether, this platform serves as a valuable tool for researchers undertaking reaction optimization programs. The model is free to use at https://photocatals.grzybowskigroup.pl/predict/.

AB - Utilizing an extensive library of literature on photocatalytic transformations, we disclose the development of a machine learning (ML) model for the recommendation of photocatalysts most suitable for reactions of interest. The model is trained on > 36 000 such literature examples and uses an architecture inspired by the Bidirectional Encoder Representations from Transformer (BERT) large language model. Under cross-validation, it can suggest the “correct” photocatalysts with ∼90% accuracy. When experimentally tested on five out-of-box reactions, this algorithm consistently suggested photocatalysts that gave yields competitive to those chosen by human researchers and frequently suggested alternative photocatalysts that are potentially more appealing than the originally selected photocatalyst. Altogether, this platform serves as a valuable tool for researchers undertaking reaction optimization programs. The model is free to use at https://photocatals.grzybowskigroup.pl/predict/.

KW - Large language models

KW - Machine learning

KW - Photocatalysis

U2 - 10.1002/anie.202514544

DO - 10.1002/anie.202514544

M3 - Article

SN - 1433-7851

VL - Early View

JO - Angewandte Chemie International Edition

JF - Angewandte Chemie International Edition

M1 - e14544

ER -

Design and experimental validation of a photocatalyst recommender based on a large language model

Abstract

Keywords

Access to Document

Fingerprint

Projects

SolOLED: TADF Dendrimers for Highly Efficient Solution-Processed OLEDs

MechanophotocatalysiS: Mechanophotocatalysis: a new frontier for solvent-minimized photocatalysis react

Allan Watson Programme Grant: Boron: Beyond the Reagent

Datasets

Design and Experimental Validation of a Photocatalyst Recommender based on a Large Language Model (dataset)

Cite this