Machine Learning Exciton Hamiltonians in Light-Harvesting Complexes.

Area of Science:

• Computational chemistry
• Biophysics
• Photosynthesis research

Background:

• Light-harvesting complexes (LHCs) are crucial for photosynthesis.
• Accurate calculation of excitonic properties is computationally expensive.
• Understanding environmental influences on LHCs is key to their function.

Purpose of the Study:

• To develop an inexpensive machine learning (ML)-based strategy for calculating excitonic properties of LHCs.
• To accurately model the impact of the natural environment on pigment interactions within LHCs.
• To validate the ML model's ability to extrapolate and predict spectral changes.

Main Methods:

• Classical molecular dynamics simulations of LHCs in their native environment.
• Machine learning prediction of the excitonic Hamiltonian for pigment aggregates.
• Training the ML model on chlorophylls from major plant LHCs.
• Testing extrapolation capabilities and environmental effect predictions.

Main Results:

• The ML strategy accurately reproduces geometrical fluctuations and electrostatic/polarization interactions.
• The model demonstrates strong extrapolation capabilities beyond the initial training set.
• Accurate prediction of environmental effects on absorption spectra was confirmed using wild-type and mutant minor LHCs.

Conclusions:

• The proposed ML-based strategy offers a cost-effective and accurate method for calculating excitonic properties of LHCs.
• This approach effectively captures the influence of the protein environment on pigment interactions.
• The model's ability to generalize suggests broad applicability in photosynthetic research.