Doctor and immunologist Derya Unutmaz utilized OpenAI's GPT-5 Pro to resolve a three-year-old research puzzle regarding how glucose affects T cell development and specialization. Unutmaz had originally performed the experiment in 2022, exposing T cells to low-glucose environments and deoxyglucose, but shelved the data after failing to explain why deoxyglucose heavily favored the production of inflammatory-response cells compared to low-glucose conditions. After uploading the data to GPT-5 Pro in late 2025, the model suggested that deoxyglucose interfered with the construction of the protein IL-2, which normally prevents T cells from becoming Th17 inflammatory cells. Unutmaz also tested GPT-5 Pro by asking it to simulate an unpublished experiment involving CD8+ T cells targeting lymphoma, and the model correctly predicted the boost in the cells' killing ability. According to Unutmaz, such AI models now function more like scientific collaborators. They can streamline literature reviews, hone hypotheses, simulate experiments, and help researchers narrow down worthwhile lab strategies, potentially saving weeks, months, or years of work. Unutmaz has also experimented with advanced tools like Codex and GPT-5.2 Deep Research to compile cancer mutation datasets and draft a T-cell-focused textbook. At the same time, the text notes that human subject matter expertise remains essential to evaluate the plausibility and significance of AI-generated insights, and that powerful biological capabilities require responsible handling and safety frameworks to mitigate potential misuse.
How GPT-5 helped immunologist Derya Unutmaz solve a 3-year-old mystery
In late 2025, immunologist Derya Unutmaz used GPT-5 Pro to resolve a three-year-old experimental puzzle regarding how glucose affects T cell development. The AI model successfully analyzed lab data and predicted experimental outcomes, illustrating how advanced AI can act as a scientific collaborator. Researchers note these capabilities can accelerate biological research while also raising safety concerns regarding potential misuse.
Why it matters
This case demonstrates how large language models can transition from information retrieval tools to active scientific collaborators capable of generating novel insights and predicting unpublished experimental outcomes, significantly accelerating biological research. However, it also highlights the dual-use safety challenges that require responsible handling by AI developers.