Programmable design of functional RNAs through multimodal generative learning
RNA engineering is bottlenecked by the tight coupling among sequence, structure, and function, leaving the field reliant on heuristics or screening. We introduce RNAverse, a multimodal generative model that learns a conditionable design grammar from RNA sequences, experimental structural probing profiles, and functional annotations. Beyond setting a new state of the art across 16 predictive…
The RNAverse model utilizes a multimodal generative approach to design functional RNAs, overcoming the limitations posed by the interdependence between sequence, structure, and function in RNA engineering. By learning a conditionable design grammar from RNA sequences, structural data, and functional annotations, RNAverse outperforms previous methods across 16 predictive modeling tasks and can generate four distinct RNA classes with minimal experimental testing.
This model can produce snoRNAs that enable premature termination codon readthrough, chimeric aptazymes for ligand-controlled gene regulation, and IGFBP3-binding aptamers with unprecedented affinity. Moreover, iterative optimization using RNAverse leads to 5' UTRs that surpass clinical benchmarks, boosting protein production in cell-based assays for gene therapies and vaccines.
The model's ability to generate functional RNAs without prior knowledge of the functional class sets a new standard in RNA engineering, establishing multimodal generative learning as a powerful tool for designing specific functional RNAs.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
