ProMORNA: AI Framework for Full-Length mRNA Design
Researchers propose ProMORNA, a multi-objective reinforcement learning framework for designing full-length mRNA transcripts directly from target protein sequences. The system uses a BART-style encoder-decoder model trained on over 6 million natural protein-mRNA pairs, then applies Multi-Objective Group Relative Policy Optimization (MO-GRPO) to balance stability, translation efficiency, and immune safety. In a case study on firefly luciferase, ProMORNA improved the in silico Pareto frontier for predicted half-life and translation efficiency over standard supervised baselines.
Key facts
- ProMORNA generates full-length mRNA transcripts de novo from a target protein sequence.
- A BART-style encoder-decoder model was trained on over 6 million natural protein-mRNA pairs.
- Multi-Objective Group Relative Policy Optimization (MO-GRPO) optimizes for multiple biological objectives simultaneously.
- Case study used firefly luciferase, excluded from training data and prompt pool.
- ProMORNA improved the in silico Pareto frontier for predicted half-life and translation efficiency.
- The framework addresses stability, translation efficiency, and immune safety.
- Published on arXiv with ID 2605.01513.
- The approach combines supervised pre-training with reinforcement learning.
Entities
—