Biomedical AI & Data Science

The Biomedical AI & Data Science Department develops, adapts, and applies AI/ML-based methods for robust, interpretable, and evidence-oriented analysis of biomedical data along a value chain, which largely aligns with the needs of the pharma and biotech industry as well as the public health sector:

• Target prioritization (better targets):
  • Rational selection of molecular target structures for future therapies
• Precision medicine (the right drug for the right patient at the right time and dose):
  • Prediction of disease risk, molecular subtype, disease progression, or treatment response at the individual patient level
• Clinical trials (better trials):
  • Simulation of (counterfactual) synthetic disease trajectories
  • Estimating causal intervention effects from observational data

To address these complex applications, state-of-the-art machine learning algorithms and artificial neural network architectures are adapted to the specific characteristics of biomedical data, including heterogeneity, sparsity, longitudinal structure, and multimodality. In addition, we develop problem-specific model architectures where existing approaches are insufficient. During the last years, our methodological work has specifically covered

• Hybrid AI/knowledge infusion, e.g., via geometric deep learning or scientific machine learning
• Generative AI and time series modeling
• Multi-modal data integration
• Causal machine learning

Work further explores under which assumptions and in which contexts AI-based predictions can be meaningfully applied to real-world clinical applications. Rather than focusing solely on predictive performance, we thus study explainability, uncertainty, consistency with biomedical knowledge, robustness, and external validity using real-world and clinical data, thereby advancing the aim of making AI medically meaningful and trustworthy.

Long-standing experience spans a broad spectrum of relevant data types in biomedicine:

• Multimodal longitudinal clinical studies, comprising, e.g.:
  • Questionnaire-based clinical assessments
  • Images
  • Molecular markers
• Real-world data:
  • Electronic health records (EHRs) and claims data
  • Digital markers: data derived from digital device technologies (e.g., gait sensors) and smartphone applications

Our services cover relevant parts of the value chain in translational biomedical research within the biotech and pharmaceutical industry. Companies are offered

• Development of tailor-made, cutting edge AI and data mining algorithms
• Applied research studies on specified biomedical use cases, focusing on the interpretability, robustness and evidential limitations of AI-based prediction models
• Systematic literature reviews
• Trainings

Our services are offered, e.g., through contract research. This way, different customer needs can be addressed while providing temporary resources and expertise to support internal projects.

• Colin Birkenbihl, Johann de Jong, Ilya Yalchyk, Holger Fröhlich (2024), Deep learning-based patient stratification for prognostic enrichment of clinical dementia trials, Brain Communications, 6(6): fcae445
• Philipp Wendland, Colin Birkenbihl, Marc Gomez-Freixa, Meemansa Sood, Maik Kschischo, Holger Fröhlich (2022), Generation of realistic synthetic data using Multimodal Neural Ordinary Differential Equations. npj Digital Medicine, 5(1): 122
• Tom Hähnel, Tamara Raschka, Stefano Sapienza, Jochen Klucken, Enrico Glaab, Jean-Christophe Corvol, Björn Falkenburger, Holger Fröhlich (2024), Progression subtypes in Parkinson’s disease identified by a data driven multi cohort analysis, npj Parkinson’s Disease, 10(1): 95
• Manuel Lentzen, Thomas Linden, Sai Veeranki, Sumit Madan, Diether Kramer, Werner Leodolter, Holger Fröhlich (2023), A Transformer-Based Model Trained on Large Scale Claims Data for Prediction of Severe COVID-19 Disease Progression, IEEE Journal of Biomedical and Health Informatics, 27(9): 4548-4558
• Colin Birkenbihl, Yasamin Salimi, Holger Fröhlich (2022), Unraveling the heterogeneity in Alzheimer's disease progression across multiple cohorts and the implications for data-driven disease modeling. Alzheimer's & Dementia, 18(2): 251– 261