AI and the World's Hard Problems · Bioengineering & synthetic biology · arXiv · 2026-10-23
ProMORNA: An End-to-End Generative Framework for Full-Length mRNA Design
Designing therapeutic mRNA requires full-length sequences that balance stability and translation efficiency, but existing methods rely on region-wise generation or costly search. The authors propose ProMORNA, a zero-shot framework that pre-trains a BART-style model on over 6 million protein-mRNA pairs and post-trains it with supervised fine-tuning and multi-objective reinforcement learning. In vitro, one candidate reached 3.33 times the reporter-activity AUC of the benchmark, though its advantage over the best GEMORNA candidate is exploratory.