RNA design across eras: from covariance models to modern generative AI

2026-05-08 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Life Sciences & Biology, Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Advanced, quick

Summary

RNA, once considered a simple intermediary, is now recognized as a dynamic molecule crucial for gene expression and cellular processes, with its functional versatility dependent on precise three-dimensional folding. Accurate RNA structure modeling is vital for biology and medicine, driving innovation in biotechnology through RNA engineering. Generative AI models, including DNA language models trained on large genomic datasets, are emerging as powerful tools for designing RNA sequences and predicting gene regulation and structure. These modern advancements build upon foundational research from the early 1990s, specifically the introduction of stochastic context-free grammars (SCFGs) and covariance models. Sean R. Eddy and Richard Durbin's 1994 paper, "RNA sequence analysis using covariance models," formalized this approach, presenting covariance models as profile SCFGs that capture sequence conservation and consensus secondary structure by identifying covariant substitutions in base-paired positions, significantly enhancing structural homology searches.

Key takeaway

For AI Scientists and Research Scientists focused on RNA biology, understanding the historical context of RNA modeling is crucial. While modern generative AI offers powerful tools for RNA design, its effectiveness is rooted in foundational probabilistic frameworks like covariance models. Incorporate both contemporary AI techniques and established methods to develop more robust and accurate RNA structure prediction and engineering solutions, ensuring a comprehensive approach to biotechnological innovation.

Key insights

RNA's functional versatility relies on 3D folding, with generative AI building on foundational probabilistic models for structure prediction.

Principles

• RNA function depends on precise 3D folding.
• Probabilistic models capture RNA structural grammar.
• Covariance models enable structural homology searches.

Method

Covariance models, a class of profile SCFGs, convert RNA family multiple alignments into probabilistic models. They capture covariant substitutions in base-paired positions to infer consensus secondary structure and sequence conservation.

In practice

• Use generative AI for RNA sequence design.
• Apply DNA language models for gene regulation prediction.
• Employ covariance models for structural homology searches.

Topics

• RNA Design
• Generative AI
• Covariance Models
• Stochastic Context-Free Grammars
• RNA Structure Prediction

Best for: AI Scientist, Research Scientist

Related on AIssential

• Orthrus: toward evolutionary and functional RNA foundation models — Orthrus is a Mamba-based RNA foundation model using contrastive learning and biological augmentations for superior RNA property prediction.
• Predicting RNA 3D structure and conformers using a pre-trained secondary structure model and structure-aware attention — trRosettaRNA2 accurately predicts RNA 3D structures and conformers by integrating pre-trained secondary structure priors and SS-aware attention.
• Why the Human Genome’s Tangled Physicality May Confound AI — The human genome's dynamic, multi-layered regulation defies simple linear models, challenging AI's current sequence-based predictive approaches.
• Proteins: A Mosaic Pattern to Rule Them All? — Proteins exhibit a conserved "Mosaic Q" pattern where amino acids cluster by chemical type, quantifiable by a specific descriptor.
• Programmable RNA translation through deep learning-driven IRES discovery and de novo generation — An AI framework unifies IRES identification, optimization, and de novo generation for programmable RNA translation.
• Order Matters in Retrosynthesis: Structure-aware Generation via Reaction-Center-Guided Discrete Flow Matching — Atom ordering in neural representations can encode chemical knowledge, improving retrosynthesis model efficiency and accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.