Builders Unscripted: Ep. 5 - Derya Unutmaz

**TL;DR:** Immunologist Derya Unutmaz shares how he transitioned from a medical background to building complex biological simulation apps using OpenAI's Codex and reasoning models, and envisions AI-driven "virtual cells" and digital twins. ## From Medicine to AI: Early Awakening Derya Unutmaz holds dual backgrounds in medicine and bioengineering. He recalls that upon graduating from medical school, he first realized biology needed artificial intelligence. "Biological systems have trillions of different components, with billions of reactions happening every second – it’s overwhelmingly complex." As early as the 1990s, he tried programming with AI, and later became even more excited by the deep learning revolution, AlphaFold, and ChatGPT. ## The Turning Point: Reasoning Models ### First experience with o1-preview In September 2024, OpenAI invited Unutmaz to test the first reasoning model, o1-preview. He gave it a prompt combining immunology and gaming: "Imagine combining a battle royale game with the immune system. How would you design a scenario where immune cells fight cancer?" o1-preview's response left him "almost emotional." He believes this marked the moment AI became inevitable for science: "When it starts reasoning, what you get becomes incredibly relevant to science." ### From GPT-4 to deeper trust Previously, GPT-4 could handle everyday tasks like literature searches and writing recommendation letters, but it couldn't answer questions like "What would the outcome of this experiment be?" o1-preview represented a leap in trust, and subsequent Pro versions and o3 further enhanced capabilities. ## A Codex Addict's Daily Routine ### From idea to app: flow cytometry analysis tool Unutmaz calls himself a "Codex addict." Every morning after coffee, his first task is to have Codex implement a new idea. He demonstrated an app built entirely with Codex: a flow cytometry analysis tool. The software processes hundreds of thousands of cell data points, performs complex operations like gating, percentage statistics, contour plots, and runs at incredible speed. "I'm not a software engineer; I'm a biomedical engineer. I can't really code – it would take me months to write a Snake game. Building these apps once felt like a dream." ### T cell receptor simulator He built an interactive app simulating the T cell receptor (TCR) signaling pathway. Users can adjust parameters like ligand quality, dose, inhibitory molecules, to simulate molecular activation, phosphorylation patterns, and predict pathway changes. "This isn't just visualizing a dataset – it's a full application you build to identify inputs and outputs for each specific cell type and scenario in the browser." ### Creative fusion of ImageGen and Codex He used ImageGen 2.0 to generate a *Nature*-style immunology atlas cover, then asked Codex to turn it into an interactive website – cells animate, clicking reveals details (e.g., memory cells, effector cells), and even simulates the effect of PD-1 blockade on cancer treatment. ## Future Vision: Virtual Cells and Digital Twins Unutmaz believes AI will have a huge impact on biology, especially in simulating complex biological systems. "When building an airplane, you simulate it first. But for biology, we don't have that." His goal is to build "virtual cells," and eventually "digital twins" of tissues and entire humans. This will require more powerful computing; he half-jokingly urges: "You'd better invest in developing this." ## Conclusion From medical school to cancer immunology, from reasoning models to everyday tools built with Codex, Derya Unutmaz is driving the deep integration of AI and biology as both a scientist and creator. His story shows how AI enables someone who isn't a software engineer to turn ideas into working applications quickly and explore the frontiers of unknown biology. **Source:** https://www.youtube.com/watch?v=4sexN3yE8xg