Fig. 1: Computational and experimental workflow for the design of artificial group I introns.

a Azoarcus ribozyme secondary structure. More conserved positions among homologous sequences of the Multiple Sequence Alignment appear in darker red. b RNAs were designed using evolutionary models (DCA, VAE, PRO), structural models (SB, SB-3D, BPR), and a combination thereof (DCA-SB). The input of evolutionary models is a Multiple Sequence Alignment. The input of structural models is ab initio predictions of secondary structure or experimental 3D contacts of Azo. Designs are generated based on probabilistic scores and/or structure scores, depending on the model. c Assay mimicking self-splicing: candidate ribozymes (black) are synthesized with an exon sequence (yellow); RNAs are mixed and incubated with the gray and red substrates; In Step 1, the ribozyme covalently attaches its 3′ yellow to the gray substrate; Then, the product, consisting of the covalently joined gray and yellow sequences, is exchanged for the red substrate; In step 2, the ribozyme ligates the red substrate onto its 3′ extremity. d Screening workflow: after computational generation (b), the ribozymes are transcribed from a DNA pool, and tested by the screening assay (c); after screening, the active ribozymes are amplified with a substrate complementary primer before being sequenced; the frequency of active sequences (carrying S2) post-assay is normalized by the variant frequency pre-assay.