RiboBridge couples RNA backbone generation to sequence design through a learned interaction topology

Authors

Ma, Z., Zheng, Y.

Abstract

Designing ribonucleic acid (RNA) molecules with prescribed structures and functions is central to RNA engineering. Backbone-first approaches generate three-dimensional geometry before assigning a nucleotide sequence, yet leave open how that geometry should be represented for downstream sequence design. Here we introduce RiboBridge, which infers a probabilistic residue-pair state from backbone geometry, recycles binary pair probabilities through a second pass and transfers binary and typed signals as soft attention biases for sequence decoding. On 550 generated backbones, pair-type decoding achieved joint refolding validity of 0.451 versus 0.335 for an architecture-matched topology-free decoder. On a matched 110-backbone intervention cohort, replacing aligned pair states with exact-length donors matched on pair-state statistics reduced joint validity from 0.464 to 0.214, showing backbone alignment matters. Pair-confidence ranking enriched the joint-validity rate from 0.451 overall to 0.596 among the top 248 before external refolding, with the same ranking signal transferring to independently generated RFDpoly and RNA-FrameFlow backbones. On 53 experimentally determined backbones, pair-coupled decoding improved secondary-structure agreement relative to residue-wise argmax. These results support a learned relational interface between sequence-free backbone generation and sequence design.

Preprint server: bioRxiv
The authors list and abstract were imported from bioRxiv on 19 Oct 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.