The results achieved in the project until now go significantly beyond the current state of the art by delivering a unified, multiscale and dynamically consistent description of how light-harvesting complexes (LHCs) regulate the balance between efficient energy capture and photoprotection across different photosynthetic organisms. The work integrates high-confidence structure prediction, long-timescale molecular dynamics, constant-pH simulations, and advanced free-energy and protonation analyses into a coherent workflow. This has enabled the reconstruction of fully pigmented, atomistic models of previously unresolved fucoxanthin–chlorophyll proteins (FCPs) in diatoms, the quantitative characterization of pH-dependent conformational landscapes in LHCs of plants and cyanobacteria, and the explicit identification of kinetically distinct macrostates associated with light-harvesting and quenched regimes for diatoms. Importantly, we demonstrate that photoprotection is not encoded in a single structural switch but in a tunable ensemble of conformations reshaped by pH, protein–protein interactions, and pigment composition.
A breakthrough is the establishment of a transferable methodological framework linking protonation equilibria to structural rearrangements and, ultimately, to excitonic and spectroscopic properties. Through the development of an approach for pKa prediction, we provide a rigorous way to decompose macroscopic pKa values into distinct local environments, thereby resolving a long-standing limitation in modeling proton-coupled conformational transitions in complex membrane proteins. By anchoring exciton dynamics and computational spectroscopy in structurally and thermodynamically consistent ensembles, the project has thus far delivered a mechanistic bridge between atomic-scale dynamics and experimentally observed quenching behavior. In parallel, the project has advanced the state of the art in ML-based exciton dynamics. By training machine-learning models on quantum chemical reference data, we have established models capable of predicting site energies, excitonic couplings, and spectral densities across thousands of molecular dynamics snapshots at a fraction of the original cost. This has enabled statistically robust sampling of exciton Hamiltonians and has opened the door to non-adiabatic molecular dynamics simulations of excited-state processes in realistic protein environments. The ML-accelerated workflows reproduce reference electronic-structure results while extending accessible timescales and system sizes well beyond conventional limits. As a result, we can now directly connect conformational variability, protonation patterns, and aggregation states to predicted absorption spectra, energy-transfer pathways, and quenching efficiencies, providing a predictive and scalable platform for computational spectroscopy of complex light-harvesting systems.
