Tsinghua and Nobel Teams Deploy Cross-Modal AI to Break RNA Bottlenecks
Aibo Bio, a Shanghai-based AI drug discovery company, has advanced its cross-modal artificial intelligence platform by deploying the ProtRNA and ProtmRNA models to address persistent data scarcity in RNA research. Spearheaded by computational biologist Xu Gang and led by CEO Ma Jianpeng, the initiative leverages foundational knowledge from protein language models to accelerate RNA sequence optimization and mRNA vaccine design. The project began in 2023 when Xu's team pivoted from traditional protein structure prediction to RNA modeling after identifying a critical bottleneck: high-quality, annotated RNA structural data remains severely limited compared to protein databases. Rather than training from scratch, the researchers adopted a cross-modal transfer learning strategy. By initializing their models with Meta's ESM-2, a highly optimized protein language model, they successfully transferred learned biochemical and evolutionary patterns to RNA sequences. ProtRNA, published in Cell Systems in 2025, demonstrated that adjusting only a fraction of parameters and utilizing a sixth of traditional training data could match or exceed performance against dedicated RNA models on standard benchmarks. Building on this architecture, the team released ProtmRNA in early 2026, specifically engineered to optimize mRNA codon regions for enhanced stability and translational efficiency. By mapping codons to their corresponding amino acid representations within the ESM-2 framework, the model significantly improved predictive accuracy for mRNA degradation, fungal expression, and overall stability, with correlation coefficients rising by up to 26.7 percent over baseline protein models. These capabilities align closely with emerging industry milestones, including the August 2026 Phase III success of a Moderna-Merck personalized mRNA cancer vaccine, underscoring the commercial relevance of precise mRNA design. Aibo Bio distinguishes its approach through a tightly integrated dry-wet laboratory closed loop. Unlike purely computational firms, the company co-locates algorithm development with expert wet-lab validation supported by Nobel laureate Michael Levitt and protein engineering authority Alan Fersht. Experimental outcomes, including failure data and intermediate states, are systematically fed back into model training. This process minimizes reliance on external contract research organizations and builds a proprietary dataset grounded in real-world biophysical constraints. While Aibo Bio's commercial pipeline currently emphasizes antibody development, the cross-modal RNA platform establishes a scalable foundation for next-generation therapeutics, including viral vaccines, gene therapy vectors, and synthetic biology applications. By decoupling AI progress from sheer data volume and embedding computational design within rigorous experimental feedback, the company is positioning itself at the intersection of algorithmic innovation and translational biopharma.
