An immunologist credits GPT-5 Pro with a breakthrough on an old mystery

GPT-5 Pro helped immunologist Derya Unutmaz solve a three-year-old T-cell mystery, showing the OpenAI model’s potential as a scientific collaborator.

Derya Unutmaz, a physician and immunologist at Jackson Laboratory and the University of Connecticut, recounts in a narrative published by OpenAI how GPT-5 Pro helped him unlock an experiment that had been on hold for three years. The question: how glucose directs the specialization of T-cells, these immune cells that fight viruses and cancer cells.

In 2022, his team had exposed T-cells either to a low-glucose medium or to a molecule, deoxyglucose, which disrupts its use. Both conditions limited available energy and should have been similar, but deoxyglucose produced significantly more inflammatory cells, a discrepancy inexplicable by energy deprivation alone. Lacking an explanation, the experiment had been set aside. By submitting the data to the model, Unutmaz says he obtained a lead: deoxyglucose would hinder the production of a protein, IL-2, which slows the shift of T-cells toward an inflammatory type, Th17, thereby removing an obstacle to this transformation. He describes the insight as remarkable and retrospectively logical, outside his immediate field. He adds that he then saw the model correctly predict the outcome of an unpublished experiment on CD8+ lymphoma-killing cells.

Unutmaz now describes these models as collaborators, useful for sifting through literature, refining hypotheses, and simulating experiments before returning to the bench. He emphasizes, however, the irreplaceable role of expertise in judging the scope of a result. OpenAI, for its part, reminds that these same capabilities necessitate safeguards, with the risk of misuse, particularly towards biological or chemical weapons, being monitored by its Preparedness framework.