Maximizing Direct Electron Transfer in Thylakoid Membranes Using Synergistic Nanomaterials Through Bayesian Optimization

Thylakoid membranes (TM) in the chloroplasts of green plants, cyanobacteria, and algae manage photosynthesis, converting solar energy into chemical energy. During photosynthesis, two photosystems of PS II and PS I split water molecules to generate electrons, protons and oxygen. The excited electron can be artificially harvested by direct electron transfer (DET) through contact with an external electrode. In the last decade, diverse nanomaterials have been extensively studied with TMs to enhance the performance bio-photoelectrochemical (BPEC) systems. The so far reported nanomaterials each focus on selective mechanisms such as enhancing electron transfer efficiency, increasing surface area for interaction, augmenting photosynthesis, and utilizing plasmon resonance energy transfer. However, enhancement mechanisms can often be conflicting and the studies on using multiple nanomaterials simultaneously with TM are lacking. Therefore, in this work, we optimized the synergistic effect of the previously proposed nanomaterials on TM through using Bayesian optimization (BO), which is a technique widely used in machine learning to optimize costly and time-limiting experiments. We selected poly(3,4-ethylenedioxythiophene):polystyrene sulfonate (PEDOT:PSS) [1], multi-walled carbon nanotube (MWCNT) [2], and far-red carbon dot (Fr-CD) [3] as our nanomaterials and set our objective function to maximizing DET of photoelectrons in TM. The optimization was conducted in a loop cycling between the BO and experimenter. Through utilizing the previously searched data, BO built a surrogate model and recommended a new search point containing information of TM wt%, PEDOT:PSS wt%, MWCNT wt%, Fr-CD wt%, and the total loading level. The weight percentages were calculated without the solvent and the total loading level of the sample was controlled by controlling the solvent amount in a fixed volume. The experimenter examined the search space by first preparing homogeneous TM-nanomaterial mixture through using Thinky Mixer set at 2000 RPM. Depending on the viscosity, the mixture was either drop-casted by 0.1ml or screen printed with a 0.1mm height on a 10mm x 10mm oxygen plasma coated indium tin oxide (ITO) electrode. The oxygen plasma coating enabled uniform contact between mixture and the ITO electrode. Under 100mW/cm² illumination, the current density output from the mixture was measured using chronoamperometry method. The measured result was fed back to the BO to update its surrogate model and calculate the next point of experiment using expected improvement (EI) policy. Through applying BO, we successfully searched the large 5-dimensional search space, and the highest-performing TM-nanomaterial mixture showed 32-folds higher current density compared to standalone TM in PBS solution.


References
[1] Yong Jae Kim, ACS Applied Energy Materials, 2023, 6, 2, 773-781
[2] JaeHyoung Yun, et al, Applied Surface Science, 2022, 575, 151697
[3] Wei Li, et al, Angewandte Chemie, 2024, 136(4), e202308951

Acknowledgement
This work was supported by the National Research Foundation of Korea(NRF) grant funded by the Korea government(MSIT) (No. 2020R1A2C3013158 and No. RS-2023-00222166).