Exploring the future of health foundation models with Tuomo Hartonen
We spoke to Tuomo Hartonen who started as a new EMBL-style group leader at the Institute for Molecular Medicine Finland (FIMM) in September. He is joining the Nordic EMBL Partnership group leader community with a mission to push the boundaries in forecasting future health trajectories.
Can you tell us a bit about your scientific background and what first drew you to your research field?
I hold master's degrees in both theoretical physics and translational medicine and earned my PhD in medical genetics from the University of Helsinki.
Theoretical physics gave me quite a strong foundation in applied mathematics and computational skills, which can be applied in any profession requiring analytical thinking and teasing out patterns from large amounts of data. On the other hand, studies in translational medicine opened my eyes to the vast spectrum of research conducted in biomedicine and helped me understand where the current and future challenges lie in bettering the health of human populations.
I have been interested in learning how things work and why the world is how it is from an early age. My first dream profession was archeologist, sparked by reading books about ancient Egypt. In high school it was already clear to me that I want to pursue a career in research, the tough decision was though whether to study history or physics, two of my favorite subjects. In the end the decision to study physics was heavily influenced by our field trip to CERN during the final year of high school. I was so impressed by the scale of that effort and how scientists there really got to push the frontiers of human knowledge!
What made you choose a Group Leader position at FIMM?
My main research interest is understanding the patterns in health data collected from entire populations and how this data can be used to improve health and make the healthcare system work more efficiently. Finland is one of the few places in the world where such high-quality data is available for research. At FIMM, there are many research groups with experience working with the Finnish health data through FinnGen and other studies. I believe at FIMM we have a rare opportunity to develop a strong research program in population-scale AI approaches for human health by integrating the existing expertise more strongly together, bringing in new collaborators for example from the ELLIS Institute Finland.
What are the main research questions your group will focus on, and why are they important?
My group develops health foundation models that integrate data from multiple sources, including diagnoses, medications, laboratory measurements and medical imaging. By modelling all data together, the foundation models can find connections from the data that might be otherwise missed. Importantly, instead of being trained to predict just one outcome, foundation models are trained to generate the next health event on an individual’s health trajectory. Here lies the biggest promise of such models: in theory they could be used to simulate future health trajectories, with vast potential in personalized medicine on individual level, and improved planning of public health interventions and strategies on the population scale.
We are also focusing on fairness and privacy issues related to foundation models, since they are one of the key challenges in translating AI into healthcare practice. Powerful AI models are very good at accidentally learning biases from the training data if such are present, meaning that in the worst case, an AI model could amplify existing health inequalities.
How do you hope to build collaborations within FIMM, the Nordic EMBL Partnership, and the wider scientific community in the region?
Our work has two clear connection points where collaboration with clinicians and biologists is highly mutually beneficial. First, our aim is not to develop methods for the sake of developing methods, but to address clinically and biologically meaningful questions. One key strength of FIMM and the Nordic EMBL Partnership is that we have people with different expertise. We have already benefited a lot in discussions with clinicians in designing our ongoing studies, and I foresee a lot more collaboration opportunities in the near future when starting to integrate AI-predicted health trajectories with omics-resources.
I am already involved in a Nordic-Baltic collaboration project that aims at testing the fairness of health foundation models across Denmark, Estonia, Finland and Sweden and developing ways to mitigate fairness issues together with input from different stakeholders.