Programmable RNA translation through deep learning-driven IRES discovery and de novo generation

2026-04-24 · Source: Nature Machine Intelligence · Field: Science & Research — Life Sciences & Biology, Mathematics & Computational Sciences, Health & Medical Research · Depth: Expert, long

Summary

A new end-to-end artificial intelligence framework addresses the challenge of precisely controlling protein expression in RNA therapeutics by unifying Internal Ribosome Entry Site (IRES) identification, evolutionary optimization, and de novo generation. The framework includes IRES-LM, an ensemble of two language models trained on 46,774 sequences, which improves IRES prediction by 15% in AUC and F1 score over existing methods and accurately identifies all 21 experimentally validated circular RNA IRESs. IRES-EA integrates an evolutionary algorithm with IRES-LM, achieving 98.4% acquired IRES functionality in 12,000 mutated sequences. Additionally, IRES-DM, a diffusion model, de novo generates novel IRES sequences with 99.3% detectable functionality from 12,000 generated sequences, producing diverse designs from natural-like to structurally conserved but sequence-divergent. This comprehensive AI framework offers a robust approach for programmable RNA translation, enhancing the molecular toolkit for biomedical discovery and RNA-based therapeutics.

Key takeaway

For AI Scientists and Research Scientists working on RNA therapeutics, this framework offers a powerful suite of tools to overcome protein expression bottlenecks. You should explore integrating IRES-LM for superior IRES identification, IRES-EA for optimizing existing RNA sequences, and IRES-DM for generating novel, functional IRES elements. This can significantly accelerate the design and development of next-generation RNA-based therapies and synthetic biology applications.

Key insights

An AI framework unifies IRES identification, optimization, and de novo generation for programmable RNA translation.

Principles

• AI can overcome complex biological design bottlenecks.
• Language models excel at sequence-function prediction.
• Evolutionary algorithms can guide targeted sequence mutation.

Method

The framework uses IRES-LM (language models) for prediction, IRES-EA (evolutionary algorithm) for targeted mutation, and IRES-DM (diffusion model) for de novo sequence generation, validated by massively parallel reporter assays.

In practice

• Use IRES-LM for highly accurate IRES prediction.
• Apply IRES-EA to optimize existing non-IRES sequences.
• Generate novel IRES elements with IRES-DM for diverse designs.

Topics

• Internal Ribosome Entry Sites
• RNA Translation Control
• Deep Learning Frameworks
• Language Models
• Evolutionary Algorithms

Code references

• a96123155/IRES_Prediction_Design

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Single-round evolution of RNA aptamers with GRAPE-LM — GRAPE-LM enables single-round RNA aptamer evolution using AI, outperforming multi-round human methods.
• RNA design across eras: from covariance models to modern generative AI — RNA's functional versatility relies on 3D folding, with generative AI building on foundational probabilistic models for structure prediction.
• Bridging three-dimensional molecular structures and artificial intelligence with a conformation description language — ConfSeq enables large language models to process 3D molecular structures by converting them into discrete token sequences.
• 🔬ESM: The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub — Scaling protein language models with vast, diverse data reveals emergent biological capabilities and enables programmable design.
• GPT-5 lowers the cost of cell-free protein synthesis — GPT-5 connected to a robotic lab significantly reduced cell-free protein synthesis costs through autonomous closed-loop optimization.
• GenoMAS: A Multi-Agent Framework for Scientific Discovery via Code-Driven Gene Expression Analysis — GenoMAS automates gene expression analysis using a multi-agent LLM framework that balances structured workflows with adaptive, code-driven problem-solving.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Nature Machine Intelligence.