Chemical Space of Molecular Nanomotors: Optimizing Photochemical Properties for One- and Two-Photon Applications

Light-driven molecular nanomotors hold promise for applications in materials science and biomedicine. Significant efforts have focused on improving their efficiency, often targeting single candidate molecules. Here, we present a systematic, data-driven approach to design nanomotors with high isomerization quantum yields for one- and two-photon applications, the latter being critical for biomedical applications requiring near-infrared light. We analyze the excited state properties of a data set of 2016 nanomotors substituted with electron-donating and electron-withdrawing (push-pull) groups. Among the top candidates, we achieved an increase in two-photon absorption strengths of up to 2 orders of magnitude compared to existing nanomotors. To ensure that the π-π* character of the excited state is preserved, which is necessary to achieve the required photoisomerization, we introduce a transition density similarity score that gauges the excited state character based on the transition. Furthermore, we benchmark three machine learning (ML) models─Kernel Ridge Regression, XGBoost, and a Neural Network─using physical and connectivity-based molecular descriptors. The excellent accuracy of our ML predictions holds promise to replace computationally costly quantum chemistry calculations in chemical space explorations.