Beyond Retinal: Machine Learn­ing Models for Photo­chem­i­cal Control in Rhodopsins

The project is devel­op­ing a machine learn­ing frame­work to accurately predict the excited states of rhodopsins. To this end, a dataset of quantum chemi­cal calcu­la­tions on retinal deriv­a­tives in protein-like environ­ments is being compiled and used for model train­ing. The models are validated and refined through the repeated synthe­sis and spectro­scopic analy­sis of specif­i­cally designed rhodopsin variants. The goal is to create a data-driven platform for the ratio­nal design of light-sensi­tive proteins and the accel­er­ated devel­op­ment of new photoreceptors.

Rhodopsins are light-sensi­tive proteins that contain a covalently bound retinal chromophore as their photoac­tive unit. Although all rhodopsins share this common core, they exhibit a wide range of photo­chem­i­cal reactions—from simple E/Z isomer­iza­tions to multi­step pathways—and thus perform differ­ent functions. This diver­sity arises from the delicate inter­play between the intrin­sic reactiv­ity of the chromophore and the modulat­ing effect of the surround­ing protein matrix on the dark, excited, and photo­prod­uct states.

A deep under­stand­ing of these inter­ac­tions is essen­tial for decipher­ing the molec­u­lar princi­ples of biolog­i­cal light percep­tion and for specif­i­cally control­ling photo­chem­i­cal reactiv­ity. Exper­i­men­tal methods such as time-resolved UV/Vis and Raman spectroscopy provide valuable data, but the ultra­fast dynamic processes compli­cate their inter­pre­ta­tion and often lead to specu­la­tive struc­ture-property relation­ships. Quantum-theoret­i­cal simula­tions of the excited state offer mecha­nis­tic insights, but are practi­cally inacces­si­ble for the large chromophore-protein complexes of rhodopsins.

This project addresses this limita­tion by devel­op­ing a machine learn­ing (ML) frame­work that describes excited states in covalently bonded systems and uses rhodopsins as a model. The project brings together the exper­tise of Prof. Dr. Klaus Robert Müller (machine learn­ing for chemistry and physics), Prof. Dr. Carolin Müller (high-quality QM/MM data and exten­sion of ML models for excited states), and Prof. Dr. Peter Hegemann (synthetic, expressed, and spectro­scop­i­cally charac­ter­ized rhodopsin deriv­a­tives). The combi­na­tion of mass-selec­tive ion soft landing and ESR STM repre­sents a ground­break­ing method­olog­i­cal advance­ment that provides a modular platform for the controlled assem­bly of arbitrary molec­u­lar build­ing blocks and their spin coupling, and can be seamlessly extended to larger biomol­e­cules (e.g., metal proteins). In the long term, an open toolset will be created for the scien­tific commu­nity that links elemen­tary surface physics with quantum infor­ma­tion and sensor technol­ogy, laying the founda­tion for the next gener­a­tion of molec­u­lar quantum simula­tors and optoge­netic tools.

Figure 2: Illus­tra­tion of the overar­ch­ing project objec­tive: Devel­op­ing excited-state machine learn­ing (ML) models to go beyond retinal model systems (left) to predict photoin­duced phenom­ena of retinal within its native protein environ­ment (colored boxes). This will be addressed by combin­ing compu­ta­tional chemistry, machine learn­ing, and spectroscopy to estab­lish a founda­tional ML framework.