• excited states
• hamiltonian learning
• light harvesting
• machine learning
• molecular dynamics
• nonphotochemical quenching

Modeling the conformational and excited-state dynamics in light-harvesting complexes requires an integrated multiscale computational approach. Molecular dynamics is needed to obtain a good conformational sampling of these complexes, while quantum chemical calculations enable the study of excited state processes. Here we have complemented these techniques with machine learning in multiple ways. First, machine learning was used to aid the interpretation of the complex conformational landscape obtained from molecular dynamics simulations of light-harvesting complexes. In addition, we have built regression models predicting the excited state properties of the embedded pigments, in a polarizable electrostatic environment, bypassing the computational burden of quantum chemical calculations.We show that machine learning techniques are powerful computational instruments for the study of these and other complex biomacromolecules.