New Research Group at BIH: AI in Healthcare and Life Sciences

Dr. Wild, what field of research do you work in, and what are your main research interests?

I work at the intersection of artificial intelligence, medicine, and the biomedical life sciences. I am particularly interested in how we can derive robust predictions about future diseases from complex health data and how these predictions can be used meaningfully to detect diseases early and, ideally, prevent them.

From a methodological perspective, my work includes foundation models, large language models, and multimodal learning. One of the key challenges for me is generalizability: medical data differ substantially between hospitals, countries, and healthcare systems. A good model should therefore not only work well on the dataset on which it was developed.

Which projects and research topics have you worked on in the past that are particularly important to you?

During my PhD, I developed AI methods for studying social networks and collective behavior. I worked across the entire experimental and computational pipeline, from data collection to the development of new analytical methods. During this time, I learned how valuable, but also how challenging, interdisciplinary research can be. Finding a common language with colleagues in biology was key to turning methodological work into research with real-world impact.

Later, I transferred this methodological focus to the life sciences. I am particularly interested in work in which we systematically predicted disease onset across many conditions simultaneously, for example based on patients’ medical histories. I find it remarkable how much AI can reveal from these data about the risk of a wide range of diseases—and how little of this potential is currently being realized in healthcare practice.

What are your plans for the future direction of your research?

I would like to develop my research in three directions. First, I want to develop models that can be transferred more effectively across different hospitals and healthcare systems. Foundation models could help overcome differences in coding systems and data structures. Second, I want to integrate different data sources within shared models. Third, I want to investigate more closely what should happen after a good risk prediction has been made. Which patients should undergo additional examinations? Which intervention is appropriate? And under what conditions does an AI-based strategy actually improve health outcomes?

Where do you see connections to the translational mission of the BIH, and how could patients benefit from your work in the future?

For me, translation means considering the entire pathway from methodological development to meaningful implementation within the healthcare system. High predictive accuracy alone is not enough. It is also necessary to determine what concrete decision follows from a prediction, whether the subsequent intervention is effective, and whether such an approach is practical and cost-effective in routine care.

AI could be particularly useful in prevention by helping to allocate limited resources more effectively. Instead of broadly applying additional examinations, they could be targeted more specifically toward people at elevated risk and for whom early intervention is likely to provide a meaningful benefit. The BIH and Charité provide ideal conditions for this: close collaboration with colleagues in clinical medicine and epidemiology makes it possible to develop and evaluate methods directly in the context of real-world healthcare questions. I want to systematically investigate this transition from prediction to prevention.