GPT-5 solves 3-year T cell mystery in minutes; suggests deoxyglucose-IL-2 mechanism, enables wet-lab validation

Immunologist Derya Unutmaz at The Jackson Laboratory for Genomic Medicine published on June 23, 2026 a detailed case study of how OpenAI's GPT-5 Pro solved a three-year stalled T cell biology puzzle in minutes. The lab had observed puzzling changes in how T cells develop when exposed to deoxyglucose, but classical analysis failed to identify the mechanism. When Unutmaz uploaded the unpublished experimental data into GPT-5 Pro, the model identified the likely mechanism: deoxyglucose interferes with construction of IL-2 (interleukin-2), a critical signaling protein.

GPT-5 Pro's analysis was not just a hypothesis—it predicted and explained previously published observations with high accuracy. When Unutmaz asked the model to simulate an existing T cell experiment targeting lymphoma, GPT-5 correctly predicted enhanced CD8+ kill capacity, despite the data being unpublished. Crucially, the model's suggestion was biologically plausible and falsifiable. The speed of analysis—from three months of unresolved data to a testable mechanistic claim in minutes—demonstrates how frontier models can compress the distance between confusing signals and actionable experiments when paired with domain expertise.

This case exemplifies OpenAI's framing of frontier AI in scientific research: not autonomous problem-solving but research workflow acceleration. The work succeeded because Unutmaz possessed the right dataset, knew which questions remained unanswered, understood why earlier analysis had stalled, and could recognize biological plausibility. GPT-5 Pro functioned as a fast, well-read collaborator that surfaces connections across literature, generates novel hypotheses, and designs follow-up experiments—shifting the bottleneck from hypothesis generation to wet-lab validation.