AI & Data Science

Our Mission: Bringing Better Treatments to the Right Patients


The research of our team has a methodological as well as an application oriented component, where method development is typically driven by specific questions arising in applications (for instance in pharma industry). Currently these applications largely cover:

  • precision medicine (the right drug for the right patient):
    • AI based modeling of disease risk, disease progression and disease subtypes
    • AI based simulation of synthetic patient data as a mechanism for privacy preserving data sharing
  • early drug discovery (better drug targets):
    • AI based drug target prioritization
    • AI methods for adverse event prediction

In addition, we have a long standing experience with applications of AI in system medicine (reverse engineering and simulation of biological networks).

To address the highly complex questions arising in our different applications a broad range of different AI and data science techniques is needed (covering neural networks, Bayesian learning, Bayesian Networks, kernel methods, boosting and others). At the same time, off-the-shelf solutions rarely provide satisfactory results. Hence, a significant proportion of our work goes into the adaptation, development and design of AI and data science techniques that are tailored to solve a particular application problem. During the last years our method developments have specifically covered

  • »Hybrid« AI: combination and integration of knowledge (e.g. in form of graphs) into machine learning models
  • (Generative) modeling of multivariate time series, including approaches to deal with missing values
  • Models that deal with multiple data modalities and biological scales.

We have historically a long standing experience with various types of -omics data, but during the last years other data types (e.g. clinical, real world evidence, bio-imaging derived features) have become more and more important.

For more information, including publications see http://abi.bit.uni-bonn.de.

Selected Publications

  • de Jong, J., Emon, M. A., Wu, P., Karki, R., Sood, M., Godard, P., ... & Fröhlich, H. (2019). Deep learning for clustering of multivariate clinical patient trajectories with missing values. Giga Science, 8(11), giz134.
  • Khanna, S., Domingo-Fernández, D., Iyappan, A., Emon, M. A., Hofmann-Apitius, M., & Fröhlich, H. (2018). Using multi-scale genetic, Neuoimaging and clinical data for predicting Alzheimer’s disease and reconstruction of relevant biological mechanisms. Scientific reports, 8(1),
    11173.
  • Benjamin Engelhardt, Maik Kschischo, Holger Fröhlich (2017), A Bayesian Approach to Estimating Hidden Variables as well as Missing and Wrong Molecular Interactions in ODE Based Mathematical Models, Journal of the Royal Society Interface, Jun;14(131). 14:20170332
  • Ashar Ahmad, Holger Fröhlich (2017), Towards Clinically More Relevant Dissection of Patient Heterogeneity via Survival based Bayesian Clustering, Bioinformatics, 33(22), 3558 - 3566
  • Y. Cun, H. Fröhlich (2013), Network and Data Integration for Biomarker Signature Discovery via Network Smoothed T-Statistics, PLoS ONE, 8(9):e73074

Interview mit Holger Fröhlich

»Die Pharma-Branche befindet sich durch den Einsatz von KI in einer Umbruchphase«

Bonn, 29.11.2018

AETIONOMY final symposium and IMI neurodegeneration initiatives