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

The paradigm is the use of computational methods for RNA sequence design.

This preprint presents RNAtranslator, a novel computational method for designing RNA sequences that bind to specific proteins. It frames the problem as a sequence-to-sequence translation task, leveraging the power of large language models. This approach represents a significant advancement within the existing paradigm of computational RNA design. It addresses limitations of previous methods, such as the need for post-generation optimization and dependence on protein-specific data, thus expanding the scope of designable targets. The evaluation shows promising results, generating novel RNA sequences with high binding affinity and stability for various proteins, including those previously considered undruggable. While experimental validation is still pending, the in silico results strongly suggest a valuable contribution within the current paradigm. Therefore, I classify this preprint as Normal Science, signifying an incremental advancement rather than a paradigm shift. No existing alternative paradigm is challenged or replaced; instead, RNAtranslator refines and extends existing computational tools for RNA design.