How GPT-5 solved an immunologist’s 3-year T cell mystery

Immunologist Derya Unutmaz faced a three-year-old puzzle about how glucose affects T cell specialization. His lab had exposed T cells to either low glucose or a glucose-like molecule called deoxyglucose, expecting both conditions to limit energy and produce similar results. Instead, deoxyglucose caused a massive shift toward inflammatory Th17 cells, while low glucose did not. The team could not explain the difference, so they shelved the experiment in 2022.

When GPT-5 Pro was released in late 2025, Unutmaz uploaded the old data and asked the model to analyze it. GPT-5 Pro suggested that deoxyglucose interfered with the construction of the IL-2 protein, which normally prevents T cells from becoming Th17 cells. The insight was just outside Unutmaz’s expertise—neither he nor his lab had seen the connection. He then tested GPT-5 Pro on a separate unpublished experiment, where the model correctly predicted that CD8+ T cells would show enhanced ability to kill lymphoma cells. Unutmaz concluded that these models have reached a point where they “truly understand.”

For Unutmaz, AI now functions as a collaborator—simulating experiments, predicting outcomes, and cutting weeks to years of lab work. He still insists that subject matter expertise remains essential to evaluate the significance of AI-generated insights. The ability to accelerate biology also raises responsible-use concerns, including lowering barriers for misuse. Unutmaz believes AI represents a shift unlike the internet or the industrial revolution, and he is using advanced tools like Codex and GPT-5.2 Deep Research to compile a T-cell-focused textbook for precision immunotherapy.