YSM Joins International Team to Develop AI for Precision Medicine

One disease can look very different across individuals. Even within the same subtype of a particular disease, patients may experience dramatically different disease progression rates, treatment responses, complications, and more. Because of this, disease models that are built on data from many people often don’t surface the biological details that might be most important for an individual and their personalized treatment plans.

In addition, when researchers build a model of one disease, it is not at all useful for understanding another.

With this challenge in mind, Naftali Kaminski, MD, Boehringer Ingelheim Pharmaceuticals, Inc. Professor of Medicine (Pulmonary) at Yale School of Medicine (YSM), began working together with researchers interested in this problem—the need for models that can both extend across disease and be biologically detailed enough for precision medicine.

The interdisciplinary, international group that they formed has now received a €16.9 million grant from the European Union’s Horizon Europe Programme to tackle this issue with artificial intelligence (AI). Their new research project, called AIRIS (Mechanism-Informed Multimodal Generative AI for Causal and Dynamical Modelling in Biomedical Research), brings together the multidisciplinary expertise of 21 research and industry partners from nine European countries, the United States, and Canada.

Over the next four years, the consortium will develop a generative AI platform that builds and reasons with mechanistic models of disease, rather than relying solely on statistical patterns. The AIRIS platform will integrate diverse biological and clinical data, such as medical scans, lab tests, genomic data and patient records, while linking molecular and cellular processes to patient-level outcomes. The goal is to design a platform that assists researchers in identifying previously unknown disease pathways and developing novel scientific hypotheses.

“We will generate tools and approaches, which could then be used to identify specific insights into disease, biomarker discovery, drug repurposing, intervention design, and more,” says Kaminski. “The goal is to greatly accelerate research on predictive and personalized medicine.”

The YSM team—the only U.S.-based team in the consortium and led by Kaminski—will head research on the specific disease domains of AIRIS focus: pulmonary fibrosis, steatotic liver disease, cardiovascular disease, chronic kidney disease, and inflammatory bowel disease in collaboration with groups from Italy, Belgium, Canada, England, Spain, Switzerland and Canada—spanning a wide range of expertise, from single cell multi-omics, through translational modeling, radiology imaging and clinical research. It will also participate in the development of AI models and tools, with Xiting Yan, PhD, associate professor of medicine (pulmonary, critical care, and sleep medicine), serving as the Yale team’s computational lead.

“This is one of the first projects, that I’m aware of, in which researchers that study these five diseases—which share mechanisms and risk factors—are actually working together,” says Kaminski. “Adding the embedded AI expertise to the team makes this a project of unprecedented interdisciplinarity and promise.”

To build disease models like the ones Kaminski and his colleagues envision, you need a lot of data. In order to really understand disease on an individual level, researchers need data across all biological scales—from molecular and cellular pathways, to single cells, tissues, and organs, to whole-body systems—and they need all of that data across time as well, as diseases emerge, change, and progress.

While researchers do collect this type of data, integrating it all into a single package that can tell the entire story of a disease in an individual is difficult, requiring considerable time, effort, and multidisciplinary expertise. And that’s where generative AI comes in.

The research team aims to build a generative AI platform that can integrate and harmonize all of the many different data types and sources required for understanding disease—something current generative AI platforms are not equipped to do—and then perform exploratory analyses on that data. The AIRIS platform will also model diseases and help identify causal mechanisms at different biological scales. And ultimately, researchers will be able to use the platform for hypothesis generation.

The group describes a scenario wherein a hypothetical researcher studying idiopathic pulmonary fibrosis uploads all of her data—spanning CT scans, lung function tests, molecular data from blood samples, clinical records, and patient-reported measures—into AIRIS, which harmonizes the data, aligns it across timelines, and flags gaps. To better understand why some patients decline more rapidly than others, the researcher works with AIRIS to develop a model that links relevant aspects of her different types of data and points to likely causal pathways. AIRIS might then suggest a biomarker that could predict rapid decline, a cancer drug that may slow disease progression in some patient groups, and a potentially relevant lifestyle intervention, delivering justifications and literature to support each. The researcher sorts through the suggestions, discarding weaker ones and refining more promising directions. She then uses AIRIS to design a new study to test the hypotheses most worth pursuing.

“It’s a full package approach to support a researcher throughout their entire project,” says Kaminski. “It will reduce the barrier to discovery and accelerate timelines so that we can begin to identify personalized medicine approaches for individuals across any number of diseases.”

In the years ahead, AIRIS seeks to transform how scientists investigate complex diseases by providing a trustworthy AI collaborator that supports every stage of the research process—moving past black-box predictions and grounding its reasoning in biological knowledge. The AIRIS consortium places a strong emphasis on transparency, robustness, explainability, bias detection and mitigation, and ethical oversight.

“We're very excited,” says Kaminski. “And while this is an international project with its own goals and experts, I’m confident this will enable many projects here at Yale as well. We have the basic sciences, the computational people, and a very large clinical operation, all of the ingredients you need to apply this approach.”

AIRIS research teams are located in Belgium, Canada, Germany, Greece, Italy, the Netherlands, Spain, Sweden, Switzerland, the United Kingdom, and the United States. The project coordinator is Christos Diou, an associate professor at Harokopio University of Athens in Greece.