GPT-5 Pro Unlocks 3-Year-Old Immunology Mystery on T-Cell Glucose Metabolism

GPT-5 Pro Bridges a Decade-Old Gap in Immunology

According to the OpenAI Blog, immunologist Derya Unutmaz, a professor at The Jackson Laboratory and University of Connecticut, encountered a critical breakthrough in late 2025 when GPT-5 Pro helped resolve a puzzle that had stalled his lab for three years. The mystery centered on a deceptively simple but consequential question: how does glucose shape the way T cells—a crucial immune-cell type—develop and specialize into distinct functional roles? This wasn’t merely an academic curiosity. T cells govern the body’s defense against cancer, viral infections, parasites, and autoimmune conditions. Understanding the molecular switches that direct their differentiation could unlock new therapeutic pathways.

The original experiment, conducted in 2022, exposed developing T cells to two glucose-restriction conditions: a low-glucose environment and exposure to deoxyglucose, a glucose analog that blocks cellular glucose uptake. Intuitively, both conditions should produce similar results—limited energy supply. Instead, the lab observed a striking divergence: deoxyglucose-exposed cells overwhelmingly differentiated into inflammatory-response cells, while low-glucose cells showed far fewer inflammatory specializations. The effect persisted even after deoxyglucose was removed. Something beyond simple energy deprivation was driving the difference, but the mechanism remained opaque.

How AI Reopened a Stalled Investigation

Unutmaz’s team shelved the experiment and pursued other priorities until revisiting it with GPT-5 Pro’s analytical capabilities. Rather than generating raw hypotheses, the AI model enabled systematic re-examination of existing data, helping the researchers identify mechanistic pathways that conventional analysis had missed. According to the OpenAI Blog, this computational assistance was instrumental in deciphering how glucose availability and glucose-like molecules activate distinct cellular signaling cascades—a distinction that energy-based explanations alone could not account for.

Why This Matters

For research teams grappling with counterintuitive biological phenomena, this case illustrates a shift in how AI augments the scientific process. Unutmaz now describes AI as inseparable from his lab’s workflow—“like taking both of your hands away, or half of your brain away,” he told OpenAI. The implications extend beyond one lab: breakthroughs in T-cell glucose metabolism directly inform immunotherapy design, potentially accelerating treatments for solid tumors, lymphomas, and inflammatory conditions. As AI-assisted hypothesis generation becomes routine in wet-lab research, teams will likely revisit archived experimental datasets with fresh computational perspectives, surfacing insights that were invisible to prior analytical techniques.