GPT-5 Pro helps immunologist solve 3-year-old T cell mystery in minutes

Immunologist Derya Unutmaz at The Jackson Laboratory published a case study Tuesday on how OpenAI's GPT-5 Pro solved a puzzle his team had been unable to crack since 2022. The question was mechanistic: how does glucose metabolism affect T cell specialization as the cells develop? Unutmaz's 2022 experiment showed that T cells in low-glucose environments behaved differently than those exposed to deoxyglucose, but the lab couldn't explain why. Three years later, he uploaded the unpublished dataset to GPT-5 Pro and asked the model to analyze the data.

GPT-5 Pro suggested that deoxyglucose interferes with IL-2 protein construction, which normally prevents T cells from becoming inflammatory Th17 cells. By blocking IL-2, deoxyglucose removes that barrier, allowing Th17cell differentiation. Unutmaz said the insight was one his lab should have reached, but hadn't, and GPT-5's mechanistic explanation proved testable: subsequent experiments confirmed the model's hypothesis. Crucially, GPT-5 had no prior knowledge of the unpublished results yet produced a biologically sound explanation.

For researchers using frontier models, the Unutmaz case is instructive: GPT-5 does not replace domain expertise or experiment. Instead, it changes the tempo of attention and hypothesis generation. The model compressed what might have taken weeks or months of literature review and brainstorming into minutes. As Unutmaz noted, models like GPT-5 now function as collaborators, not oracles—they narrow the search space, but human validation and wet-lab follow-up remain essential. This pattern will repeat in any field where clean data, domain expertise, and disciplined validation converge.