A generalized and efficient approach for complete mRNA design improves translation, stability and specificity.

Area of Science:

• Molecular Biology
• Bioinformatics
• Machine Learning

Background:

• Messenger RNA (mRNA) properties like expression, immunogenicity, and stability are sequence-dependent, offering potential for optimization.
• Current machine learning approaches struggle with unified, end-to-end mRNA design due to complex component rules and challenges in out-of-distribution tasks.

Purpose of the Study:

• To develop a unified algorithmic approach for complete mRNA design.
• To leverage machine learning and diverse RNA data for optimizing mRNA molecules.
• To address limitations in adapting AI tools for novel mRNA design tasks.

Main Methods:

• Introduction of Design by Integrated Gradients (DIGs), an alteration to integrated gradients for mRNA design.
• Integration of predictive neural networks with diverse RNA datasets (transcriptomic profiling, CLIP-seq).
• Application of the DIGs framework for complete, model-informed mRNA sequence design.

Main Results:

• Demonstration of complete model-informed mRNA design using the DIGs technique.
• Identification of previously underexplored rules for assembling functional mRNA components.
• Successful design of mRNA sequences in out-of-distribution settings.

Conclusions:

• The DIGs framework unifies neural network strengths with extensive RNA data for robust mRNA design.
• Designed mRNA sequences exhibit significantly improved translational capacity and cell-type-specific expression.
• This approach facilitates the creation of high-performance mRNA molecules for specific applications.