yakRNA Design: A semantic multimodal RNA composer

Authors

Pinpin, L. N., Khan, Y. A.

Abstract

Like proteins, RNAs have been a target for exploitation to generate synthetic molecules that can adopt enhanced and novel functions. High-throughput assays, combined with meticulous biochemistry, have led to the generation of some artificial RNAs but the ability to algorithmically program RNAs with an intended function remains difficult. While generative models have revolutionized protein design, RNA design remains challenging due to the dearth of 3D RNA structures. Here we present You Always Know RNA (yakRNA) Design, a frontier language model that can simultaneously reason over sequence, biological function, consensus sequence, and secondary structure to generate functional RNAs. yakRNA Design is not trained on any 3D structural information but rather co-learned over a corpus of semantically labeled sequences that we designed. yakRNA Design, without any human intervention, fine-tuning, sequence optimization, scoring or selection, zero-shot designed 17 (out of 84 total designs) synthetic RNAs that were able to efficiently induce ribosomes to change their reading frames during elongation at rates comparable to or higher than most natural sequences. One of these efficient synthetic RNAs was found to have no identity to any sequence in the known universe, demonstrating how a model with semantic understanding of RNA has rich generative capabilities.

Preprint server: bioRxiv
The authors list and abstract were imported from bioRxiv on 24 Apr 2026.

Stats

1-terrible, 9-excellent. How would you rate this preprint? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.