[2S01m] 10. ASPIRE - Part 4

11:01 AM - 11:21 AM JST(2:01 AM - 2:21 AM UTC)

[2ASPR-05] AI-enabled rhodopsin design for artificial photosynthesis

○Wei Huang¹ (1. University of Oxford (UK))

Keywords:

aritificial photosynthesis,rhodopsin,AI,machine learning

We developed an AI-guided design pipeline that generated non-natural microbial rhodopsins with properties not known in nature, marking a significant advance in programmable design of synthetic biology. This approach produced a new class of blue-light-absorbing rhodopsins, expanding the usable light-harvesting spectrum beyond the range of previously reported proton-pumping rhodopsins. The pipeline combined a novel algorithm for sequence generation, a machine-learning regressor for ranking candidates, and a Markov-based structural plausibility model for filtering implausible designs, enabling efficient in silico discovery. Experimental validation showed that the top variants were substantially blue-shifted and stable, showing high proton-pumping activity. These designed rhodopsins also enhanced the growth of Cupriavidus necator on formate as the sole carbon source, confirming their functional value in a microorganisms. These AI-designed rhodopsins are the most blue-shifted proton-pumping rhodopsins reported so far and they have not been identified in nature. This study demonstrates the novelty and power of AI not only to engineer existing biological functions, but also to create entirely new light-harvesting modules for artificial photosynthesis with potentially higher solar energy utilisation efficiency.