Synteny Awarded Largest Allocation in UKRI AIRR ‘AI for Science’ Programme to Build a Sovereign Probabilistic World Model for Goal-Directed Immunological Discovery
London, United Kingdom – Synteny Labs (synteny.ai), a frontier AI life sciences company training foundation models of biological systems, today announced it has been awarded 1,000,000 GPU hours on Isambard-AI through UK Research and Innovation’s (UKRI) AIRR “AI for Science” supercomputing competition.
The 1,000,000 GPU-hour allocation is the largest allocation granted under this call. At the published rate of £4.50 per Isambard-AI GPU hour, the award represents approximately £4.5 million (~$6 million USD) in advanced AI compute infrastructure.
Isambard-AI is powered by NVIDIA GH200 Grace-Hopper superchips, representing the most advanced AI compute architecture currently deployed in UK sovereign infrastructure.
Synteny will use the compute allocation to scale its development of a probabilistic world model of adaptive immunity, a multimodal foundation model designed to discover the functional principles governing immune receptor-antigen interactions.
The system is designed to move beyond static biological prediction toward closed-loop, goal-directed discovery, integrating calibrated uncertainty with experimental feedback. This work underpins Synteny’s broader mission to enable the design of programmable protein therapeutics, transforming adaptive immunity into a computationally navigable engineering substrate.
“This substantial compute enables Synteny to train a probabilistic world model for biological discovery. By training our world model and building virtual systems, we are advancing autonomous scientific discovery and strengthening strategic capabilities central to the UK’s AI agenda,” said Lilly Wollman, Co-founder & CEO of Synteny.
The AIRR “AI for Science” programme is a core component of the UK’s strategy to accelerate AI-enabled scientific discovery by providing access to sovereign, large-scale compute systems, including Isambard-AI and Dawn. These systems deliver frontier-scale performance and memory bandwidth specifically optimised for training large foundation models, placing them among the most powerful AI compute resources in Europe.
Awards were made through a competitive assessment process prioritising technical ambition, scientific impact, and alignment with national priorities in medical research and AI-driven innovation. Synteny’s selection for the maximum allocation reflects both the scale of its proposed foundation model training and the strategic importance of autonomous systems in engineering biology.
Over the past 12 months, Synteny has trained its flagship interactome foundation model and generative models on TCR-pHLA interactions, achieving SOTA on challenging protein engineering tasks. However, modelling the combinatorial and structural complexity of adaptive immunity requires substantially larger architectures and extensive convergence validation.
The GPU-hours of this scale support training runs of the final architecture and the essential prototyping and large-scale ablations needed for convergence.
The project is exclusively research and pre-clinical in scope. It does not involve personal or clinical data and does not support patient-level decision-making. All model outputs are subject to human oversight and experimental validation, with controlled access to prevent misuse.
Synteny is a frontier AI company training foundation models of biological systems to enable scientific discovery. By integrating machine learning with large-scale experimental platforms, Synteny develops closed-loop AI systems capable of learning biological principles and guiding experimentation toward defined objectives. The company’s mission is to unlock programmable protein therapeutics, with a focus on TCR and HLA-directed therapeutics across oncology and autoimmunity, through self-learning, foundation-scale modelling of adaptive immunity.