Design of diverse, functional mitochondrial targeting sequences across eukaryotic organisms using variational...

Area of Science:

• Mitochondrial biology
• Metabolic engineering
• Bioinformatics

Background:

• Mitochondria are crucial for cellular energy production and metabolism.
• Effective mitochondrial targeting sequences are limited, hindering metabolic engineering and disease treatment.
• Passenger proteins significantly influence protein localization efficiency.

Purpose of the Study:

• To design novel mitochondrial targeting sequences using a Variational Autoencoder.
• To validate the functionality and utility of these designed sequences in eukaryotic organisms.
• To explore the evolutionary origins of dual-targeting sequences via latent space interpolation.

Main Methods:

• Utilized a Variational Autoencoder (VAE) for de novo design of mitochondrial targeting sequences.
• Performed in silico analysis to predict peptide functionality and mitochondrial targeting features.
• Experimentally characterized artificial peptides in four eukaryotic organisms.
• Applied latent space interpolation for evolutionary analysis.

Main Results:

• Generated peptides showed a high predicted functionality rate (90.14%) for mitochondrial targeting.
• Demonstrated proof-of-concept applications, including increased 3-hydroxypropionic acid titers via pathway compartmentalization.
• Significantly improved delivery of 5-aminolevulinate synthase by 1.62-fold and 4.76-fold.
• Identified potential evolutionary pathways for dual-targeting sequences.

Conclusions:

• Generative artificial intelligence is a powerful tool for designing functional mitochondrial targeting sequences.
• The developed sequences offer practical applications in metabolic engineering and disease therapy.
• This approach advances fundamental understanding of mitochondrial biology and protein targeting.