How GPT-5 helped immunologist Derya Unutmaz solve a 3-year-old mystery
TL;DR
Immunologist Derya Unutmaz had a T-cell puzzle open since 2022: low glucose and deoxyglucose were expected to behave similarly, but the lab results diverged. GPT-5 Pro suggested that deoxyglucose disrupted IL-2 production. With that brake weakened, many more T cells could specialize into inflammatory Th17 cells. Unutmaz also tested GPT-5 Pro on an unpublished CD8+ experiment against lymphoma cells. The model correctly predicted the stronger cancer-cell killing effect.
Nauti's Take
This is one of the better GPT-5 examples because it does not stop at magical framing. GPT-5 Pro does not replace the lab or an immunology background; it shifts where humans get unstuck, from noticing a data pattern to forming a plausible mechanism.
That is where AI can become scientifically useful. The PR packaging is predictable, but the core is strong: models matter when they point researchers toward testable next experiments.
Briefingshow
The important part is not that GPT-5 Pro produced an answer, but that it surfaced a mechanistic link a specialist lab had not prioritized for years. In biology, that is where leverage sits: sharper hypotheses, fewer dead ends, and faster selection of experiments worth running. Without experts like Unutmaz, the model output is still only a hypothesis.
Sources