Training mRNA Language Models Across 25 Species for $165

OpenMed built an end-to-end protein AI pipeline covering structure prediction, sequence design, and codon optimization, training four production models across 25 species in 55 GPU-hours for $165. The pipeline uses ESMFold for protein folding and ProteinMPNN for sequence design. For codon optimization, CodonRoBERTa-large-v2 achieved a perplexity of 4.10 and a Spearman CAI correlation of 0.40, outperforming ModernBERT. Maziyar Panahi contributed to the work, with CodonJEPA listed as an upcoming model on the project roadmap.