Immunologist Derya Unutmaz, professor at The Jackson Laboratory and the University of Connecticut, revived a stalled 2022 study after uploading its data to GPT-5 Pro. The model parsed results from a glucose‑deprivation experiment on developing T cells and produced the insight that finally explained the three‑year‑old puzzle about how sugar influences immune specialization, using natural‑language prompts to query patterns and generate hypotheses.

The team had compared T cells grown in low‑glucose media with cells exposed to deoxyglucose, a glucose analog that blocks energy production. Contrary to expectations, deoxyglucose‑treated cells skewed toward inflammatory Th17 phenotypes. GPT-5 Pro traced the effect to suppressed production of IL-2, a cytokine that normally restrains Th17 differentiation, revealing the missing regulatory link. This mechanistic clue directed subsequent assays that confirmed the pathway.

Unutmaz now treats the model as a research partner, using it to scan new literature, simulate experiments and prioritize hypotheses, cutting months of bench work. He cautions that domain expertise remains essential to vet AI‑generated mechanisms. OpenAI’s Preparedness Framework governs responsible deployment, aiming to prevent misuse while accelerating precision immunotherapy development. This workflow cut cycles dramatically. The collaboration has already yielded faster insights into cancer‑fighting T cells.