UTRGAN: learning to generate 5' UTR sequences for optimized translation efficiency and gene expression

Area of Science:

• Synthetic Biology
• Bioinformatics
• Molecular Biology

Background:

• The 5' untranslated region (UTR) of mRNA is critical for controlling protein expression levels and stability.
• Optimizing UTR sequences is essential for high and stable protein production in synthetic biological systems.
• Existing UTR sequences are often patented, necessitating the development of novel, high-performance alternatives.

Purpose of the Study:

• To develop a computational model for generating novel 5' UTR sequences with improved properties.
• To optimize generated UTR sequences for enhanced target gene expression, ribosome load, and translation efficiency.
• To provide a publicly available tool for designing synthetic UTRs for biological applications.

Main Methods:

• Utilized a Generative Adversarial Network (GAN) framework named UTRGAN to generate 5' UTR sequences.
• Implemented an optimization procedure to enhance predicted protein expression, ribosome load, and translation efficiency.
• Validated generated UTR sequences by comparing their predicted properties and experimental translation rates against known UTRs.

Main Results:

• UTRGAN-generated UTRs demonstrated up to five-fold higher predicted expression and a 34-fold higher predicted translation efficiency compared to initial sequences.
• Sequences exhibited increased similarity to known regulatory motifs, including internal ribosome entry sites and Kozak sequences.
• In vitro experiments confirmed higher translation rates for TNF-α protein using UTRGAN-designed UTRs compared to the high-capacity Beta Globin 5' UTR.

Conclusions:

• UTRGAN provides an effective AI-driven method for designing synthetic 5' UTRs with significantly enhanced translational properties.
• The model's ability to mimic natural UTR characteristics and optimize for specific expression metrics offers a valuable tool for synthetic biology.
• The open-source release of UTRGAN and its dataset facilitates further research and application in optimizing protein expression systems.