Fig. 3: Organism labeling and workflow for in vivo evaluation of generated artificial mitochondrial targeting sequences (AMTSs).

a Selection of AMTSs for experimental characterization. AMTSs are generated through the VAE, encoded via the pre-trained UniRep model, and annotated with an organism label based on proximity to the cluster center, defined as the mean UniRep embedding of MTSs in a proteome, and using k-Nearest Neighbors. b Scarless DNA assembly of AMTSs utilizing PfAgo/AREs. Plasmid backbones are cleaved using PfAgo and phosphorylated guides, creating sticky ends. Annealed DNA sequences, encoding for AMTSs, are then ligated upstream of fluorescent reporter proteins using T4 DNA ligase. Final constructs are verified through Colony PCR, demonstrating high assembly fidelity. c Functional validation of positive control MTSs. Plasmids harboring MTS-fluorescent protein constructs are transformed into the host organism, followed by verification of mitochondrial localization using fluorescence microscopy. Mitochondria in HEK293 cell lines are visualized with MitoTrackerTM Orange CMTMRos, whereas mitochondrial targeting in N. benthamiana is confirmed through protein colocalization with scCOX4-mCherry[2]. Scale bar: 12 μm. b, c Each experiment was repeated three times independently with similar results. Source data are provided as a Source Data file. a, b Created in BioRender. Zhao, H. (2025) https://BioRender.com/wszxpbe.