Improvement of Exciton Hamiltonians for Photosynthetic Reaction Centers by Bayesian Optimization

We implemented the Bayesian optimization (BO) method to modify exciton Hamiltonians for the photosynthetic reaction centers (RCs) to achieve better reproduction of the experimentally measured optical and circular dichroism (CD) spectra. The Python library "PHYSBO" was applied to modify the conventional exciton model for the structure of the photosystem I RC of a cyanobacterium Thermosynechococcus elongatus (T. elongatus) as an example. The BO method modified the exciton Hamiltonian, which was precalculated by the Poisson-TrESP/charge density coupling method. The method reproduced experimental properties of absorption and CD spectra, including the long-wavelength absorption bands of so-called Red-Chlorophyll a (Red-Chl a) unique for photosystem I. The candidate molecules of Red-Chl a are shown to be situated at low site-energy sites as revealed by the mutation experiments. The tuning of parameters by the BO method effectively improved the exciton Hamiltonians for photosystem I and the other type-I RCs.