Designing organic photoredox catalysts as a fragment-based puzzle solved by reinforcement learning
Abstract
Reinforcement learning (RL) is a promising strategy for identifying optimal solutions in vast search spaces. We have developed a framework in which the design of organic photoredox catalysts (OPCs) is deliberately formulated as a fragment-based puzzle that is well suited to RL. In this framework, an RL agent sequentially selects and orders molecular fragments and attempts to identify fragment combinations that afford high rewards, as calculated based on the similarity between the properties of the generated molecules and those of 4CzIPN, a well-known OPC. In addition to the rediscovery of a previously reported high-performance OPC, experimental validation of the proposed candidates revealed a TADF emitter with redox potentials similar to those of 4CzIPN, as well as an OPC with superior catalytic activity in several photoreactions.
Related
- Detonation Nanodiamond-Supported Ru(II) Poly(pyridine) Complexes as Efficient and Recyclable Photoredox Catalysts
- Designing Porous Boroaluminate Catalysts via Non‐Hydrolytic Ether Elimination
- Machine Learning-Guided Strategy Selection and Condition Prediction for Photoredox Catalysis.
- Fragment-Based Representation Learning for Polymers and Heterogeneous Formulations
- Active Learning for Fragment-Based Kinase Inhibitor Discovery using Docking
BibTeX
@misc{noto2026designing,
title = {{Designing organic photoredox catalysts as a fragment-based puzzle solved by reinforcement learning}},
author = {Naoki Noto and Mayu Hirakata and Mikito Fujinami and Yusuke Masuda and Ryosuke Kojima and Yasunori Matsui and Hiroshi Ikeda and Hiromi Nakai and Masaya Sawamura and Susumu Saito},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010145/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010145/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Designing organic photoredox catalysts as a fragment-based puzzle solved by reinforcement learning
AU - Naoki Noto
AU - Mayu Hirakata
AU - Mikito Fujinami
AU - Yusuke Masuda
AU - Ryosuke Kojima
AU - Yasunori Matsui
AU - Hiroshi Ikeda
AU - Hiromi Nakai
AU - Masaya Sawamura
AU - Susumu Saito
PY - 2026
DA - 2026/10/09
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010145/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010145/v1
AB - Reinforcement learning (RL) is a promising strategy for identifying optimal solutions in vast search spaces. We have developed a framework in which the design of organic photoredox catalysts (OPCs) is deliberately formulated as a fragment-based puzzle that is well suited to RL. In this framework, an RL agent sequentially selects and orders molecular fragments and attempts to identify fragment combinations that afford high rewards, as calculated based on the similarity between the properties of the generated molecules and those of 4CzIPN, a well-known OPC. In addition to the rediscovery of a previously reported high-performance OPC, experimental validation of the proposed candidates revealed a TADF emitter with redox potentials similar to those of 4CzIPN, as well as an OPC with superior catalytic activity in several photoreactions.
ER -