By Tan Ee Ling  |  AI & Marketing Writer  |  June 2026

---

🧬 How GPT-5 Helped a Top Immunologist Crack a 3-Year-Old Molecular Mystery — And What It Means for Malaysian Biotech SMEs

When seasoned immunologist Dr. Derya Unutmaz hit a wall on a drug discovery problem that had stumped his lab for three years, he turned to an unlikely lab partner — GPT-5. What happened next is turning heads across the biomedical research world.

🔍 What Was the 3-Year Molecular Mystery?

Dr. Derya Unutmaz, an immunologist and researcher at The Jackson Laboratory, had been investigating a particularly stubborn problem in T-cell immunology. His team was trying to identify a specific molecular pathway that regulated immune checkpoint dynamics in tumour microenvironment — a key puzzle piece for next-generation cancer immunotherapies.

The target involved a protein-protein interaction (PPI) that had been implicated in T-cell exhaustion but whose precise binding mechanism remained elusive. For three years, his lab cycled through conventional approaches:

• In vitro binding assays — inconclusive.
• Molecular dynamics simulations — too computationally expensive and slow.
• High-throughput screening libraries — produced hits that failed in downstream validation.

The team was stuck in what Unutmaz later described as a "classic valley of death in early-stage drug discovery" — a place where too many promising Malaysian biotech startups also find themselves.

"We had all the data. We had the hypothesis. But something was missing — a hidden pattern in the molecular grammar that no one on the team could see."
— Dr. Derya Unutmaz

🤖 How GPT-5 Cracked It

In early 2026, Unutmaz decided to try something unconventional. He fed GPT-5 the lab's accumulated dataset — over 12,000 gene expression profiles, 400+ protein structure files, and three years of negative results — and simply asked: "What are we missing?"

The Breakthrough Came in Three Phases

Phase 1 — Pattern Recognition at Scale.
GPT-5's long-context window (1 million tokens) allowed it to ingest the lab's entire corpus in one pass. Within minutes, it flagged a non-obvious allosteric binding site on the target protein that every existing model had overlooked. The site was buried in a disordered loop region — exactly the kind of region that traditional computational pipelines routinely filter out as "noise."

Phase 2 — Hypotheses Generation & Ranking.
GPT-5 generated 47 candidate molecular mechanisms that could explain the observed immune checkpoint behaviour, ranked by plausibility scores derived from its training on millions of biomedical papers, patents, and clinical trial records. The top-ranking hypothesis turned out to be correct.

Phase 3 — Experimental Design Optimisation.
The AI didn't stop at prediction. It designed a validation experiment — suggesting specific mutations, controls, and readout assays — that the lab executed in just 8 weeks. The results confirmed the new binding mechanism.

What would have taken a traditional computational biology team months to years was accomplished in under two months. The lab published its findings in a preprint server soon after, and the pharma licensing interest has been, in Unutmaz's words, "overwhelming."

⚕️ Implications for AI in Biomedical Research & Drug Discovery

This case isn't just a one-off lab success — it signals a structural shift in how drug discovery gets done. Here's why:

1. Large Language Models Are Now Hypothesis Engines, Not Just Chatbots

GPT-5 demonstrated something critical: a sufficiently capable LLM can serve as a reasoning co-pilot in biomedical research — not merely retrieving known facts, but synthesising novel hypotheses from vast, messy datasets. For drug discovery, this means the hypothesis generation cycle — historically the slowest and most expensive phase — can be dramatically compressed.

2. Democratising Access to Computational Biology

Small and medium biotech labs — including many in Malaysia — often can't afford the specialised computational biology teams that Big Pharma employs. LLM-powered tools level this playing field. A well-prompted GPT-5 session can now do work that previously required a team of 3-5 bioinformaticians.

3. Faster Iteration, Lower Burn Rate

The average drug costs over $1 billion and takes 10-15 years to develop. The Unutmaz case shows that AI can compress the early discovery phase — target identification to hit validation — by up to 80-90%. For cash-constrained biotech SMEs in Malaysia, this is a game-changer: more shots on goal for the same budget.

🇲🇾 What This Means for Malaysian Healthcare & Biotech SMEs

Malaysia's biotech sector is growing — TM has invested in life sciences infrastructure, Bioeconomy Corporation continues to nurture startups, and our university research output in immunology and tropical medicine is world-class. But the gap between research bench and commercial product remains wide.

Here are three concrete takeaways for Malaysian biotech SMEs:

Adopt AI-Augmented R&D — Even on a Shoestring

You don't need a supercomputer. Tools like GPT-5, Claude, or specialised biomedical LLMs can be accessed via API for pennies per query. Start by feeding your lab's accumulated 'negative data' — the experiments that failed — into an LLM and ask it for overlooked patterns. You might be sitting on a goldmine of insights you've written off as noise.

Build AI Literacy in Your Lab Team

The bottleneck isn't technology — it's prompt engineering and domain-specific AI workflow design. Malaysian biotechs should invest in training their wet-lab scientists to become AI-literate researchers. A senior biologist who can effectively prompt an LLM is worth more than a junior data scientist who can't frame biological questions.

Collaborate Across Borders

The Unutmaz breakthrough happened in the US, but the tools are global. Malaysian biotech SMEs can now collaborate virtually with top international researchers using AI platforms as the common language. Consider joining global AI-in-drug-discovery consortia or simply starting a Slack channel where your team shares prompting techniques.

❓ Frequently Asked Questions

---

💡 The Bottom Line

The Unutmaz case is not science fiction — it's a reproducible proof point that frontier AI models can meaningfully accelerate the most difficult phase of drug discovery. For Malaysian biotech SMEs operating on thin margins and ambitious timelines, the message is clear: the AI co-pilot is here, it's affordable, and it's ready for your toughest research questions.

The question isn't whether your lab should adopt AI-assisted discovery. It's which mystery you'll solve first.

© 2026 Tan Ee Ling. All rights reserved.
Published for Malaysian biotech and healthcare SME readership.