Boehringer Ingelheim AI & Data Science Fellowship Program

The Else Kröner Medical Scientist School’s ClinBrAIn: Artificial Intelligence for Clinical Brain Research projects work at the interface of clinical brain research and artificial intelligence. The joint initiative of clinical groups from the University Hospital and the Hertie Institute for Clinical Brain Research and AI research groups (the Cluster of Excellence: Machine Learning, New Perspectives for Science, the Tübingen AI Center and the Bernstein Center for Computational Neuroscience) will address various diseases of the nervous system and use approaches from robust, explainable and probabilistic machine learning.

The Horizon Europe funded European Cohorts of Patients and Schools to Advance Response to Epidemics (EuCARE) project contributes to SARS-CoV-2/COVID-19 research across Europe, Kenya, Mexico, Russia and Vietnam. We aim to deliver recommendations for optimized clinical management and treatment with the support of strong immuno-virological and Artificial Intelligence (AI) components.

The interdisciplinary project HALTA_Long_COVID develops advanced artificial intelligence methods tailored to classify COVID comorbidities and symptom courses while effectively addressing patient heterogeneity. In contrast to existing models, we employ Federated Continuous Time Latent Transition Analysis accelerated by amortized inference approaches such that large data sets with complex individual patterns can be analyzed. Furthermore, we use federated representation learning to gain insights to identify patient groups with similar patterns. We apply these novel techniques to analyze data from different national and international cohorts as well as data from two consortia that have been funded by the European Union ensuring the necessary large sample sizes. Combining the developed methods and the available data, HALTA allows tackling diverse research questions related to the post-COVID syndrome while facilitating the use of new data via standardized open-source pipelines, leading to a deeper understanding of symptom progression and comorbidities.

Our project Medical Data Privacy and Privacy-preserving ML on Healthcare Data (MDPPML) is funded by the German Aerospace Center (DLR).

Privacy preserving machine learning is one of the research topics that our group is interested in. We are not only developing machine learning methods for precision medicine, but we also aim to address the privacy concerns occurring during the application of these methods in real life scenarios. We try to protect the privacy of the data of patients involved in the training/testing of machine learning methods without sacrificing the performance. More specifically, we develop frameworks for collaborative training of kernel-based machine learning methods. In order to show the efficacy of our frameworks, we utilize them on precision medicine problems, such as predicting the coreceptor usage of HIV based on V3-loop sequences.

Privacy-Preserving Analytics in Medicine (PrivateAIM) develops the next-generation federated machine learning (ML) and data analytics platform for the Medical Informatics Initiative (MII), where analyses come to the data instead of data coming to the analyses.
Methods with innovative privacy models for multimodal data will be integrated into a distributed infrastructure, which can easily be adopted by the Data Integration Centers (DICs). A thorough consideration of challenges on the intersection of technology and law, the development of concepts and documents for operation by hospital IT departments in coordination with ethics committees, information security and data protection officers will facilitate translation into practice, as well as ensure availability and sustainability.

Our project Certifications and Foundations of Safe Machine Learning Systems in Healthcare is funded by Carl Zeiss Stiftung.

The Centre of Innovative Care (Zentrum für Innovative Versorgung) is a joint project of university hospitals in the German Federal State of Baden-Württemberg: Freiburg, Heidelberg, Mannheim, Tübingen and Ulm. Our project goal is to enhance patient care through the use of new, innovative digital medicine technologies like mobile health apps. We aim to further include the patient into the treatment process, to streamline the work of clinicians and caregivers and empower scientific use of the collected data. The multimodal integration of processes and data will allow us to better understand diseases and to develop new therapy and prevention concepts in the context of given structures and institutes across Baden-Württemberg.

On behalf of the German Federal Government a General Long Covid/PASC-Data Model is developed. Under the Lead of the German Foundation for General Medicine and Family Medicine  (Deutsche Stiftung für Allgemeinmedizin und Familienmedizin, DESAM), we are working on this Data Model.

Common Action against HIV/TB/HCV across the Regions of Europe (CARE) aimed at analysing and addressing the human immunodeficiency virus (HIV), tuberculosis (TB) and hepatitis C (HCV) epidemics across Europe and Russia, by establishing new networks and consolidating previous cooperative actions between the European Union and the Russian Federation. The project generated novel research findings based on existing biological material and data previously collected by the participating partners, as well as on new samples and data gathered during the project.

The project Extending deep kernel approaches for better prediction and understanding of ADME phenotypes and related drug response was funded by the Innovation Fund program of the Cluster of Excellence Machine Learning: New Perspectives for Science.

PRIMO (Personalized medicine for tailored cancer therapies) had the goal to improve personalized treatment concepts for cancer patients with advanced molecular and computational analysis methods. This joint project of the FZI Research Center for Information Technology, NMI Natural and Medical Sciences Institute at the University of Tübingen and Hahn-Schickard was funded by the Ministry of Economic Affairs, Labour and Housing of Baden-Württemberg (WM BW). Aiming at improving the efficacy in breast cancer treatment by advanced individualization and digitization, a workflow for multi-view data integration was established and a prototype of a personalized breast cancer treatment decision support system was developed. With additional molecular patient data like advanced protein profiling and ex-vivo tumor cell models that were developed in this project, an integrative multi-omic precision oncology analysis based on Machine Learning methods was developed to gain further insights on patient-specific cancer subtyping and to assist clinical treatment decisions.

The TTU Hepatitis project, funded by the German Center for Infection Research (DZIF), aimed to optimize the use of new directly acting antiviral (DAA) based therapies for chronic hepatitis C by defining algorithms that arrive at individualized treatment regimens.

XplOit, funded within the i:DSem initiative of BMBF, aimed to develop a software plattform to support data integration, model development and model validation with medical data using semantic tools. As an application scenario XplOit collected and annotated data from patients who had undergone hematopoietic stem cell transplantation at the University Hospitals Homburg and Essen. Our part of the project was to use this data to develop models predicting critical events after transplantation, for example CMV infection / reactivation, graft-versus-host disease or overall survival. Ideally, these predictive models have the potential to warn physicians early and improve clinical care.