Machine Learning Optimization of Non-Kasha Behavior and of Transient Dynamics in Model Retinal Isomerization

Designing a model of retinal isomerization in rhodopsin, the first step in vision, that accounts for both experimental transient and stationary state observables is challenging. Here, multiobjective Bayesian optimization is employed to refine the parameters of a minimal two-state-two-mode (<i>TM</i>) model describing the photoisomerization of retinal in rhodopsin. The optimized retinal model predicts excitation wavelength-dependent fluorescence spectra that closely align with experimentally observed non-Kasha behavior in the nonequilibrium steady state. Further, adjustments to the potential energy surface within the <i>TM</i> model reduce the discrepancies across the time domain. Overall, agreement with experimental data is excellent.