GPT-5 Pro in Immunology: Solving T-Cell Specialization Mysteries

GPT-5 Pro has demonstrated the ability to solve complex, long-standing biological mysteries by analyzing experimental data and generating mechanistic insights that elude human experts. In a case study from OpenAI, immunologist Derya Unutmaz used the model to resolve a three-year-old puzzle regarding T-cell specialization, proving that the model can function as a scientific collaborator capable of predicting unpublished experimental results.

Resolving the T-Cell Glucose Mystery

GPT-5 Pro identified a specific protein interference mechanism that explained why deoxyglucose caused T cells to specialize into inflammatory-response cells at a higher rate than low-glucose environments.

In 2022, Professor Derya Unutmaz and his team observed an anomaly: T cells exposed to deoxyglucose (a glucose-like molecule that disrupts energy and protein construction) produced significantly more inflammatory-response cells than those in a low-glucose environment, despite both conditions limiting energy. The researchers were unable to explain this difference at the time and shelved the experiment.

Upon analyzing the data with GPT-5 Pro in late 2025, the model suggested that deoxyglucose specifically interfered with the construction of the protein IL-2. Because IL-2 typically prevents T cells from becoming Th17 inflammatory-response cells, the removal of this protein barrier by deoxyglucose accelerated the specialization process. This insight allowed the researcher to retrospectively understand the experimental results that had previously been inexplicable to his lab.

Predictive Capabilities in Lymphoma Research

GPT-5 Pro successfully predicted the outcome of an unpublished experiment involving CD8+ T cells and lymphoma, demonstrating a capacity for simulation rather than simple pattern matching from existing internet data.

To test the model's predictive power, Unutmaz provided GPT-5 Pro with the parameters of an experiment he had already conducted but not yet published. The experiment focused on the ability of CD8+ T cells to target and kill lymphoma cells. GPT-5 Pro correctly predicted that these cells would show an enhanced ability to kill the lymphoma cells, confirming the model's ability to simulate biological processes based on provided data.

Impact on Scientific Workflow and Research Acceleration

AI models like GPT-5 Pro are transitioning from simple tools to scientific collaborators that streamline the research lifecycle in several key ways:

• Literature Review and Hypothesis Generation: Models can process hundreds of weekly academic papers to identify unanswered questions and refine hypotheses.
• Experimental Simulation: By simulating experiments and predicting outcomes, researchers can narrow down which physical lab experiments are most worthwhile, potentially saving weeks, months, or years of manual labor.
• Data Compilation: Advanced tools such as Codex and GPT-5.2 Deep Research are being used to compile large-scale cancer mutation datasets and generate research materials, including T-cell-focused textbooks for precision immunotherapy.

The Role of Human Expertise and Safety

While AI accelerates discovery, subject matter expertise remains essential for evaluating the plausibility and significance of AI-generated insights. A researcher without specialized knowledge in immunology would not have been able to determine if the insight regarding IL-2 was scientifically meaningful.

Furthermore, OpenAI notes that the ability to accelerate biological research also introduces risks. The potential for bad actors to use these capabilities to design biological or chemical weapons is addressed through OpenAI's Preparedness Framework, which tracks risks and implements safeguards against capabilities that could cause severe harm.