Optimizing 5'UTRs for mRNA-delivered gene editing using deep learning

Area of Science:

• Biotechnology
• Molecular Biology
• Bioinformatics

Background:

• Messenger RNA (mRNA) therapeutics offer revolutionary potential but lack sequence optimization methods for increased expression.
• Efficient translation of mRNA is crucial for therapeutic efficacy, yet sequence design remains a challenge.

Purpose of the Study:

• To design 5' untranslated regions (UTRs) for efficient mRNA translation using deep learning.
• To evaluate the performance of designed 5'UTRs in supporting gene editing enzyme expression and activity.

Main Methods:

• Utilized deep learning, gradient descent, and generative neural networks for 5'UTR sequence design.
• Performed polysome profiling on randomized 5'UTR libraries across three cell types to generate training data.
• Experimentally tested designed 5'UTRs with mRNA encoding megaTAL gene editing enzymes in two cell lines and for two gene targets.

Main Results:

• 5'UTR performance demonstrated high correlation across different cell types.
• Designed 5'UTRs significantly enhanced gene editing activity.
• While editing efficiency showed cross-cell type and cross-target correlation, optimal UTR performance was context-specific.

Conclusions:

• Model-based sequence design holds significant potential for optimizing mRNA therapeutics.
• Deep learning approaches can effectively guide the design of high-performing mRNA sequences.
• The developed method facilitates the creation of more potent and efficient mRNA-based therapies.