Date: Spring, 2025 Interview Transcript (Original: In German)

How did you end up working as a postdoc in AI & Biomedicine in a lab at Harvard University?

I’ve always been fascinated by understanding relationships in our world and modeling them mathematically. My academic path began with studies in Computer Science and Applied Mathematics at ETH Zurich and continued with a PhD in Probabilistic Machine Learning at the Swiss AI Institute IDSIA in Lugano.

However, during my rather theoretical doctoral work, I realized that I wanted to align my research more closely with concrete, meaningful applications—with the goal of having a real impact on people’s lives. That’s why I accepted a postdoc position in Zurich around two years ago, where I applied machine learning models in various clinical applications: from cancer research to organ transplantation matching, rheumatic diseases, and delirium in intensive care to diabetes. A particular focus was on longitudinal models and optimal therapy decisions.

At the beginning, it was a steep learning curve for me—I had hardly any medical expertise. But the projects fascinated me so much that I immersed myself in all of them. I then wanted to learn even more about biology and medicine, so I specifically looked for a research institution that uses AI and mathematics to model complex processes in biology and medicine. I’ve now found this opportunity in Boston—more precisely in an interdisciplinary team at Harvard University in the Department of Biostatistics and the Department of Stem Cell Research as well as at the Dana-Farber Cancer Institute. There, I now work together with biologists, doctors, mathematicians, and AI experts on the medicine of tomorrow.

👉 If you would like to get in touch with Manuel Schürch: Instagram-Profile; LinkedIn-Profile 👈

What does a typical day look like for an AI & ML researcher at Harvard?

Very varied and versatile—and strongly dependent on the current project phase. On some days I write a paper, program new AI models, or analyze new results and data. On others I read current literature from the AI or biomedical community, attend seminars, give talks at conferences or workshops, or meet with colleagues for project discussions.

Since our field of research is highly interdisciplinary, regular meetings with mathematicians, clinicians, biologists, or bioinformaticians are also on the agenda. This is particularly exciting—but also challenging, as each discipline has its own “language” and brings its own perspectives.

Especially here in Boston, there are countless opportunities for new ideas, projects, and collaborations—and the chance to work with highly motivated, brilliant minds from all over the world. This makes research here not only productive but also incredibly inspiring.

Can you tell us more about your current projects?

At the moment, I’m working on several projects—three of which currently form my main focus. What particularly motivates me is working on applications with real impact. I don’t just want to write theoretically interesting but ultimately impractical papers. Instead, I’m drawn to research that has the potential to actually improve the lives of patients.

1. AI foundation models for immunotherapies in cancer treatment

In this project, we are developing an AI model based on large clinical and genomic datasets from thousands of cancer patients. The goal is to predict the individually best possible immunotherapy—that is, the treatment that promises the best survival chances for seriously ill patients. The model has enormous potential: On the one hand, it can serve as a decision-making aid for physicians, and on the other, it can provide new hypotheses for the development of personalized drugs. It combines cutting-edge machine learning technology with concrete clinical relevance.

2. Transplantation medicine: AI for organ and stem cell transplantation

Another project deals with improving organ and stem cell transplantation, especially kidney transplants. It’s about developing AI algorithms that enable personalized and optimal allocation based on genetic compatibility. Instead of considering only a few parameters such as blood type or HLA matching, we analyze thousands of genetic variables. This way, donors and recipients can be optimally matched—with the goal of increasing success rates and extending lifespan after a transplant. This project can literally give many people a new life.

3. Virtual cells: AI-driven modeling of biological systems

The third project is a bit more abstract, but at least as fascinating: We’re working on an “AI Virtual Cell”—an AI-based system that simulates cellular processes over time, including hypothetical interventions, to predict how each individual cell will develop when a drug is applied. You can imagine it like this: Every cell has its own genetic information and behaviors. It changes through processes such as cell division or mutation and interacts with other cells mainly via proteins. We model these cells as autonomous AI agents—comparable to mini-LLMs or ChatGPTs—that can autonomously communicate with each other and learn from each other.