Nabla Bio's JAM-2 AI generates antibody candidates for human trials in 2 years
In brief
- JAM-2 model generated high-affinity antibody candidates against 26 disease targets
- Over 50% of candidates met developability criteria without optimization
- Nabla Bio plans first-in-human trials within one to two years
JAM-2 Model Performance
Nabla's JAM-2 model demonstrates substantial improvements over conventional antibody design methods. For nearly half of the 26 targets, JAM-2 candidates achieved picomolar to single-digit nanomolar affinities—binding strengths that typically require extensive optimization cycles in traditional drug development.
More than 50% of the JAM-2 candidates met developability criteria without any optimization. This efficiency matters. It compresses timelines and reduces the iterative work that usually consumes months or years in preclinical research.
The model also showed strong epitope precision, routinely hitting 30-70% of user-defined epitopes. Precision matters in antibody design because it determines whether a therapeutic actually engages its intended target without off-target effects.
Targeting GPCRs and Beyond
GPCRs are the target class for roughly a third of all FDA-approved drugs, making them a high-value focus for drug developers. JAM-2 achieved up to an 11% success rate for direct on-cell binding against GPCRs—a class historically difficult for antibody-based approaches.
Some of the AI-generated candidates demonstrated the ability to activate cellular signaling pathways. This functional capacity suggests the antibodies don't just bind targets; they can trigger the biological responses needed for therapeutic effect.
Technical Architecture
JAM-2 employs test-time scaling, an approach inspired by the reasoning methods used in systems like OpenAI's ChatGPT. Rather than committing to a single prediction, test-time scaling allows the model to explore multiple solution paths and refine its output, improving both accuracy and binding affinity predictions.
Partnership and Timeline
Nabla expanded its partnership with Takeda in a deal announced in October 2025. The terms included upfront payments in the double-digit millions, with potential milestone payments exceeding $1B. This financial backing and validation from a major pharmaceutical partner reinforce Nabla's position as a credible player in AI-driven drug discovery.
The company's timeline—first-in-human trials within one to two years—represents an aggressive but plausible acceleration compared to traditional antibody development, which typically spans five to seven years from design to clinical testing.
