Understanding NLO Switches with Inverse Molecular Design and Explainable Machine-Learning

Eline Desmedt, Mercedes Alonso and Freija De Vleeschouwer

With traditional direct molecular design approaches, only small parts of the chemical compound space (CCS) are explored hampering the discovery of new functionalized molecules for potential nonlinear optical applications. Recently, we took up the challenge to apply an inverse molecular design algorithm, called the Best-First Search (BFS) algorithm, to discover promising functionalized hexaphyrins-based molecular switches with high nonlinear optical (NLO) contrasts.[1-3] Using BFS, we were able to significantly enhance the NLO contrast of the individual [26]- and [30]hexaphyrin-based redox switches (26R ⟶ 28R and 30R ⟶ 28R). In the process, an extensive dataset was collected.[2,3] Statistical analysis revealed how each type of functionalization affects the NLO responses of the [26]- and [30]-hexaphyrins, and in which aspects both redox switches differ. First, certain combinations of core-modifications and meso-substitutions have a synergistic effect on the NLO response of the [30]hexaphyrins, which is not observed for the [26]hexaphyrins. In addition, both redox switches benefit from push-pull combinations, but they each prefer a different arrangement in terms of electron-donating and electron-withdrawing groups. Next, we extended our search to multistate switches of the type 26R ⟶ 28R ⟶ 30R, where we aimed at designing switches with a similar NLO response for both ON states.[3] By visualizing the CCS, we observe that our best-performing switches are found in regions shared by high-responsive [26]- and [30]-hexaphyrins. Finally, to further understand why these two hexaphyrins behave so differently, we used machine-learning (ML) models with explainable ML techniques. We discovered that the prominent contribution of orbital and charge-transfer-based features differs between the two hexaphyrin-based switches.[4]

Figure 1.

1. E. Desmedt, T. Woller, J.L. Teunissen, F. De Vleeschouwer, M. Alonso, Front. Chem. 2021, 9, 786036.
2. E. Desmedt, D. Smets, T. Woller, M. Alonso, F. De Vleeschouwer, Phys. Chem. Chem. Phys. 2023, 25, 17128.
3. E. Desmedt, L. Serrano Gimenez, F. De Vleeschouwer, M. Alonso, Molecules 2023, 28, 7371.
4. E. Desmedt, M. Jacobs, M. Alonso, F. De Vleeschouwer, in preparation.

 Eline Desmedt