The RNA world hypothesis suggests that catalytic RNA molecules could be the starting point of the intricate history of genetic inheritance and variations that shaped Life as known today. Supporting this idea, the Azoarcus ribozyme from the group I intron (GII) family has been identified as one of the few RNA ribozymes that can autocatalytically self-reproduce. However, the realism of this RNA-centric origin of life hinges on the frequency of such RNA autocatalytic self-reproducers within the sequence space—the ensemble of all possible nucleotide sequences representing RNA molecules of a given length.
In this work, we harnessed this sequence space with generative models and tested these designs experimentally for catalysis. We designed thousands of RNA molecules computationally using multiple strategies and showed that : i) designing RNAs using an ML generative model can robustly predict experimentally active molecules, ii) the broader exploration of the sequence space requires a hybrid approach that incorporates the molecular structural information.
Our results shed new light on the relationship between molecular structure and catalytic function, helping us understand the landscape of RNA self-reproducers and the plausibility of the RNA-world hypothesis.

​