Speaker: Sophia Wesely, Leipzig University

Understanding excited-state energy landscapes is central to designing photoreactive molecules for applications in bioimaging, optogenetics, or photopharmacology. The green fluorescent protein (GFP) chromophore, p-hydroxybenzylidene-2,3-dimethylimidazolinone anion (HBDI⁻), is a model system for excited-state processes, but presents significant challenges for theory due to long-lived excited states, competing relaxation pathways, and the prohibitive cost of accurate quantum chemical simulations[1]. These factors have traditionally made systematic exploration of functional group modifications intractable. In this work, we overcome these limitations by combining our recently developed equivariant machine learning model for excited states, X-MACE [2], with nonadiabatic molecular dynamics. Trained on thousands of organic chromophores, X-MACE generates potential energy surfaces transferable across chemical space and electronic states with minimal additional data. This approach enables, for the first time, efficient screening of HBDI⁻ derivatives to assess how structural changes modulate excited-state topologies and dynamics. Our results uncover design principles for tuning photophysical properties through targeted substitutions, demonstrating the potential of machine learning to unlock excited-state landscape exploration at scale.

[1] List, N.H., Jones, C.M. & Martínez, T.J. Chemical control of excited-state reactivity of the anionic green fluorescent protein chromophore. Commun. Chem. 7, 25 (2024)
[2] Barrett, R., Ortner, C., Westermayr, J., Transferable Machine Learning Potential X-MACE for Excited States using Integrated DeepSets. preprint arXiv:2502.12870 (2025).