RiboGRAM learns multi-scale non-coding RNA grammar for structural and regulatory inference

For centuries, protein-coding sequences have been deciphered using the triplet genetic code, but the arrangement of information within non-coding RNA sequences has remained elusive. A recent study has unveiled a multi-scale "grammar" governing non-coding RNA organization, paving the way for advanced structural and regulatory inference.

Researchers have identified two key features of non-coding RNA grammar: multi-scale motif usage and positional gradients from 5' to 3'. By developing a model called RiboGRAM, scientists sought to learn this organizational framework across nine organisms. The findings revealed that non-coding RNAs exhibit scale-dependent divergence from coding sequences, presenting directional U-rich and AU-rich gradients.

The RiboGRAM model combines multi-scale sequence encoding with contextual modeling to organize RNA families. It reveals motif-associated and pair-like patterns across different layers of organization. Remarkably, non-coding RNA pretraining without structural supervision outperformed coding or messenger RNA pretraining in base-pair recovery. The model's representations enable alignment-free diffusion-based 3D modeling of short RNAs.

In a human-specific analysis, RiboGRAM identified position-resolved motif organization across protein targets and cell types. The study found sequence-preference shifts associated with reduced cross-cell transferability, highlighting the importance of non-coding RNA grammar for accurate structural and regulatory inference.

By establishing a learnable grammar of non-coding RNA sequence organization, this research marks a significant step forward in understanding these critical yet enigmatic molecular entities.