GPT-5 Pro Solves Immunological Puzzle

In a significant leap for GPT-5 Pro scientific research, OpenAI's latest model has helped immunologist Derya Unutmaz unravel a three-year-old mystery concerning T-cell development. This breakthrough, detailed in a recent OpenAI News publication, underscores the AI's potential to augment human expertise in complex biological fields.

Unutmaz, a professor at The Jackson Laboratory and the University of Connecticut, turned to GPT-5 Pro in late 2025 to revisit an perplexing experiment from 2022. The puzzle centered on how glucose affects the specialization of T cells, critical immune cells that combat cancer, autoimmune diseases, and infections.

AI Unlocks a Biological Secret

The initial experiment investigated T cells exposed to either low-glucose environments or deoxyglucose, a molecule that disrupts glucose utilization. Researchers anticipated similar outcomes, expecting limited energy to dictate T-cell behavior. However, deoxyglucose exposure dramatically increased the production of inflammatory-response T cells, a result not mirrored in the low-glucose group.

This discrepancy stumped Unutmaz and his team. Uploading the unresolved data into GPT-5 Pro, the model quickly suggested that deoxyglucose interfered with the construction of the IL-2 protein. IL-2 is known to prevent T cells from becoming Th17, an inflammatory-response cell type.

GPT-5 Pro's insight illuminated that deoxyglucose essentially removed a barrier, allowing T cells to more readily become Th17 cells, explaining the previous experimental anomaly. Unutmaz described the AI's finding as a "remarkable insight" that was outside his and his lab's immediate expertise.

The model’s predictive capabilities were further demonstrated when it accurately simulated an unpublished experiment on lymphoma-targeting T cells. This confirmed its deep understanding, beyond merely recalling internet data.

Accelerating Scientific Discovery

Unutmaz now considers AI models like GPT-5 Pro indispensable collaborators. They can process vast amounts of scientific literature, identify unanswered questions, and refine hypotheses. This significantly reduces the time required to pinpoint the most promising experiments.

By simulating experiments and predicting outcomes, AI can trim weeks, months, or even years from research timelines, accelerating biological discovery. However, human subject matter expertise remains crucial for evaluating the plausibility and significance of AI-generated insights.

The rapid advancements in AI for scientific applications, including tools like Codex and GPT-5.2 Deep Research for compiling cancer mutation datasets and generating research materials, highlight a transformative era. Unutmaz views this period as unprecedented, surpassing even the impact of the internet or the industrial revolution, and feels privileged to be part of this wave of discovery.