GPT-5 Pro Resolves 3-Year T Cell Mystery via Gene Analysis

On June 23, 2026, OpenAI and Dr. Derya Unutmaz at The Jackson Laboratory for Genomic Medicine published an analysis detailing how GPT-5 Pro resolved a three-year biological puzzle. The model analyzed legacy experimental data to identify an elusive mechanism affecting human T cell development. For researchers managing extensive historical datasets, the case study demonstrates how frontier models can synthesize multi-disciplinary data to generate actionable scientific hypotheses.

T Cell Metabolism and Glycosylation

T cells serve as critical components of the human immune system, directly responsible for fighting cancer and regulating autoimmune diseases. Since 2022, the research team had struggled to explain how 2-deoxy-D-glucose (2-DG) influenced T cell specialization. Their physical experiments consistently showed an unexpected increase in the proinflammatory Th17 cell subset when cultured with 2-DG. Human specialists could not isolate the exact mechanistic pathway driving this developmental shift.

Dr. Unutmaz supplied GPT-5 Pro with legacy gene expression datasets and a core experimental figure. Within minutes, the model proposed a novel connection. It hypothesized that 2-DG was not merely restricting energy, but interfering with N-linked glycosylation, an essential protein folding process.

The ability to process raw visual data alongside structured genomics datasets proved critical. Instead of relying solely on text-based literature reviews, the AI evaluated the actual experimental outputs. It mapped the gene expression variations across different age groups, recognizing a structural pattern that human researchers missed.

The model concluded that this interference reduced IL-2 signaling, fundamentally altering the developmental path of the T cells. The lab immediately recognized this connection as a viable, testable hypothesis that had previously eluded specialists due to its cross-disciplinary nature.

Model Capabilities and Ecosystem Scaling

The Jackson Laboratory discovery relies heavily on the reasoning capabilities introduced in the GPT-5 series. Models capable of bridging discrete scientific disciplines can uncover structural relationships across massive datasets. OpenAI recently formalized this capability with the launch of GPT-Rosalind, a domain-specific model targeted directly at biochemistry, genomics, and protein engineering.

OpenAI is concurrently upgrading the broader model family to support larger data ingestion:

Model Release	Focus Area	Key Capability
GPT-5 Pro	General Reasoning	Cross-disciplinary synthesis
GPT-Rosalind	Life Sciences	Domain-specific biochemistry reasoning
GPT-5.6	Scale and Efficiency	1.5M token capacity

The ongoing rollout of GPT-5.6 introduces a 1.5M token context window alongside a 10-15 percent improvement in token efficiency. Expanded context windows directly support this class of scientific research by allowing models to hold thousands of research papers, raw tabular data, and visual figures in a single prompt block.

Industry Implications

Unutmaz received the OpenAI Pro AI Award for his application of LLMs in accelerating research timelines. The model functioned as an analytical collaborator, synthesizing months of conventional data review into minutes of compute time. The underlying approach of using AI to isolate novel biological pathways serves as a foundational component in the OpenAI-Novo Nordisk strategic partnership, which applies similar architectures to drug discovery and supply chain optimization.

If you process legacy scientific data, large language models now possess the requisite reasoning depth to identify missing variables. Supplying frontier models with raw experimental figures and structured datasets can surface hidden interactions before committing resources to physical lab experiments.