Excitation Energy Transfer in a Photosynthetic System with more than 100 Million Atoms

点击翻译按钮获取中文摘要

Light-harvesting protein-pigment complexes of plants, bacteria and algae are key players in the conversion of sunlight into stable forms of chemical energy during the process of photosynthesis. Chlorophyll, bacteriochlorophyll and bilin molecules are the main pigments present in those complexes which absorb solar light and pass it on to other pigments. Aim of those complexes is to transport the excitation energy to reaction centers where the charge separation takes place for further processing.The objective of this project is to enable the simulation of these excitation energy transport processes at an atomistic level in a full photosynthetic chromatophore vesicle from a purple bacterium with more than 130 million atoms. While the in-silico construction of this chromatophore has been a research topic by itself, we rely on this model and aim at carrying out quantum mechanics/molecular mechanics (QM/MM) simulations on this entire cell organelle along a molecular dynamics trajectory. Until now, it was only possible to carry out very preliminary QM/MM simulation studies on this extremely large model system due to the sheer size of the model with more than 2.000 pigments. If we were to use traditional simulation techniques from the field, accurately determining the excitation energies and their transport over a longer time period would be computationally unfeasible, even on large high performance computing installations.Therefore, we propose to enrich multi-scale QM/MM simulations, which are a state-of-the-art technology from computational biophysics, by multi-fidelity machine learning (ML) models. QM/MM combines the accuracy of QM calculations with the speed of MM simulations. Similarly, multi-fidelity ML models incorporate information from many, fast to compute low accuracy quantum chemical calculations with a few very accurate quantum chemical simulations. In our novel multi-scale multi-fidelity approach, the highly accurate but fast to obtain multi-fidelity ML models will replace the QM excitation energy calculations for the pigment molecules, which make up the vast majority of the computational runtime in the QM/MM simulation. Thereby, we expect to overcome the currently existing computational barrier.To actually carry out the full scale excitation energy transport simulation, fundamental advances have to be made in methods in computational biophysics and machine learning. A joint novel methodology of QM/MM simulation using multi-fidelity ML will be the first outcome of this project. The second outcome will be the complete simulation that will ultimately allow us to answer fundamental questions in biophysics such as: Do excitation energies depend on the local neighborhood of the pigment-protein complex? Is an energy funnel present which drives the excitation energies towards certain parts of the chromatophore? What are the relevant transfer times in the chromatophore between various parts of the system?