GPT-5 Pro helped explain a 3-year T cell experiment mystery
Immunologist Derya Unutmaz’s lab ran an experiment in 2022 to understand how glucose affects the way T cells develop. A low-glucose setting and a setting with were expected to act in a similar way, because both made it harder for the cells to use glucose. The results were very different: T cells exposed to became Th17 inflammatory-response cells in much larger numbers, and the effect continued even after was removed.
The lab could not explain the result as a simple lack of energy, so the experiment was set aside for three years. suggested that may have disrupted of a protein called IL-2. IL-2 can help stop T cells from becoming Th17 cells, so weakening that signal could explain why pushed so many cells in that direction.
also correctly predicted the outcome of an unpublished -related CD8+ T cell experiment. Unutmaz now uses , Codex, and GPT-5.2 for literature review, choosing which experiments to test, organizing large cancer mutation , and drafting research materials.
Key points
- gave a possible explanation for a T cell result that had puzzled the lab for three years.
- The key idea was that may block IL-2, which can stop T cells from becoming Th17 cells.
- also predicted an unpublished result involving CD8+ T cells attacking cells.
- The same is now being used for literature review, experiment planning, cancer mutation data, and research drafts.
- The result still depended on expert review; the model’s answer was not treated as automatically true.