Did AI solve a 3-year-old medical mystery? A surprising experience from a practicing immunologist

Imagine this: a massive puzzle that hadn’t been solved despite three years of scrutiny. What if a “smart assistant” you keep by your side could suddenly provide a clue to solve this challenge that thousands of scientists couldn’t? This is a surprising story that recently unfolded in the medical field.

Professor Derya Unutmaz, an immunologist at The Jackson Laboratory for Genomic Medicine, felt like he had been lost in a maze of complex genetic data for the past three years [Source 2, Source 4]. However, he recently gained a powerful new colleague: AI. Specifically, OpenAI’s GPT-5 and GPT-5.5 models.

Why is this important?

This case shows that AI has evolved beyond simple sentence generation into a “research partner” that solves complex scientific challenges faced by humanity. The AI we use in our daily lives not only drastically shortens researchers’ work hours but has also captured relationships between genes that humans had missed [Source 7]. This suggests the possibility that AI could exponentially accelerate research into intractable diseases or drug development in the future.

Understanding it easily

Analyzing genomic data is like searching for a single line of text containing specific information among tens of thousands of books in a library. The data Professor Unutmaz handled was massive, consisting of 62 samples and about 28,000 genes [Source 2, Source 7].

To put it simply: if the existing method was a human researcher holding a magnifying glass and checking one by one, AI is like a “super-reader” that reads the contents of tens of thousands of books simultaneously and identifies the correlations between them. GPT-5 analyzed patterns in the data that humans hadn’t considered and proposed a “mechanistic hypothesis” (an assumption about the operating principle) [Source 5]. Based on the hypothesis provided by the AI, actual experiments were conducted in Professor Unutmaz’s laboratory, and surprisingly, the AI’s prediction was confirmed to be true [Source 5].

How was the research conducted?

Professor Unutmaz and his research team had struggled for three years analyzing the data without finding a breakthrough. Then, they introduced the latest AI models, GPT-5 and GPT-5.5 Pro, into their research process. They went through the process of meticulously scanning the data through AI and identifying unknown T-cell (a type of immune cell) subpopulations.

Collaboration occurred not just by asking the AI questions, but by setting the direction for data analysis and verifying whether the hypotheses proposed by the AI were biologically sound. As a result, the AI performed pattern discovery work that could have taken human researchers months or even years in a short period.

Where do we stand?

Current state-of-the-art models like GPT-5.5 Pro show excellent performance in analyzing vast amounts of academic data and experimental results [Source 4, Source 7]. However, there is a point to be cautious about. AI is merely a “smart partner” that helps researchers set hypotheses and suggests directions. The final scientific validation—the process of proving and interpreting the hypothesis proposed by the AI in an actual laboratory environment—must remain the responsibility and task of human researchers.

What will happen in the future?

The changes AI will bring to the scientific community in the future are even more anticipated. As the hypotheses proposed by AI become more sophisticated, the time required for drug development will be drastically reduced. In particular, the speed of finding cures for complex diseases like cancer or autoimmune diseases will become much faster. The era has arrived where the latest AI models are now firmly listed as essential “experimental equipment” in laboratories, beyond being simple convenience tools.

MindTickleBytes’ AI Reporter View

For scientists floundering in a sea of complex data, AI has become a source that hands them a compass. AI’s analytical capabilities now complement human cognitive limitations, rapidly expanding the realm of scientific discovery that we have yet to reach.

References

1. How Omio is building the future of conversational travel | OpenAI URL: https://openai.com/index/omio/
2. Introducing GPT-5.5: OpenAI’s New Class of Intelligence for Real Work and Agentic AI - Kingy AI URL: https://kingy.ai/ai/introducing-gpt-5-5-openais-new-class-of-intelligence-for-real-work-and-agentic-ai/
3. Introducing GPT-5.5 | OpenAI URL: https://openai.com/index/introducing-gpt-5-5/
4. Early experiments in accelerating science with GPT-5 | OpenAI URL: https://openai.com/index/accelerating-science-gpt-5/
5. Early science acceleration experiments with GPT-5 URL: https://cdn.openai.com/pdf/4a25f921-e4e0-479a-9b38-5367b47e8fd0/early-science-acceleration-experiments-with-gpt-5.pdf
6. ChatGPT Images 2.0, Workspace Agents & GPT-5.5 | April 2026 URL: https://jasonpollakmarketing.com/2026/04/23/chatgpt-images-2-workspace-agents-gpt-5-5-openai-april-2026/
7. Start Chatting with GPT - Use AI - OpenAI ChatGPT-5 Unlimited URL: https://www.bing.com/aclick?ld=e8-NveUz3ci4M3pWQMuC5_jjVUCUz5zPAztNtahDabSuADiFmDgId3BPFsFl_iEWWhZAfGSrRKF6xI9Nap81YgfqjZGIAxg0hixBJkqWk8mGOHm7uAV4I9GMncNH4dFBzJNpWVUCWDadq-VJQJ2QZ_xola7Epre7p-WMypMg–mK3Bra2fiU6EoY9L-Unq54T6-MfSV2tlTmJyIlyPt5nI7JTnpWw&u=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&rlid=a9be66ebc40d147003e20cd1075f7949