RNAtranslator: Modeling protein-conditional RNA design as sequence-to-sequence natural language translation

Area of Science:

• Biochemistry
• Computational Biology
• Molecular Biology

Background:

• Protein-RNA interactions are crucial for gene regulation, splicing, RNA stability, and translation.
• RNA is a promising therapeutic agent for targeting proteins, including those considered undruggable.
• Designing RNA sequences for selective protein binding is challenging due to vast sequence space and limitations of current methods.

Purpose of the Study:

• To develop a novel computational method for designing RNA sequences that selectively bind to target proteins.
• To overcome limitations of traditional in vitro selection and computational approaches for RNA design.
• To enable efficient RNA design for a wide range of proteins, including those lacking prior RNA-interaction data.

Main Methods:

• Introduced RNAtranslator, a generative language model framing protein-conditional RNA design as a sequence-to-sequence translation problem.
• Trained the model on large-scale datasets to learn a joint representation of RNA and protein interactions.
• Directly generated binding RNA sequences for given protein targets without post-generation optimization.

Main Results:

• RNAtranslator generated RNA sequences with natural-like properties and high novelty.
• The designed RNA sequences exhibited enhanced binding affinity compared to existing methods.
• The model successfully designed RNA for diverse protein targets, including those with no available RNA-interaction data.

Conclusions:

• RNAtranslator offers an efficient and broadly applicable approach for designing RNA sequences with specific protein-binding capabilities.
• This method facilitates the development of novel RNA-based therapeutics and advances synthetic biology applications.
• The generative model approach overcomes key challenges in RNA sequence design, expanding possibilities for targeting proteins previously considered undruggable.