AI & Data Science Fellowship Program

The partnership between the University of Tübingen and Boehringer Ingelheim is setting the next milestone in the development of new perspectives in AI and data science for healthcare. With this joint AI & Data Science Fellowship Program aimed for postdoctoral candidates, academic research will be linked with application related topics of the pharmaceutical industry.

The genetics of human populations are entangled through shared ancestry, which introduces correlations in medical and genome wide association studies. This project aims to enhance the accuracy of medical studies by leveraging large-scale genealogical analysis. Therefore we will infer huge ancestral recombination graphs that describe the local ancestry of the individuals as a sequence of thousands of genealogical trees along their genomes. We will develop machine learning techniques to process these graphs and classify individuals based on the inferred genetic relationships and disease states within cohorts. By integrating fine-scale genealogical analysis with medical research, we aim to improve cohort selection and the robustness of medical study outcomes.

Job advertisement

Efficiently identifying biocatalysts is crucial for developing sustainable chemical processes, particularly in industries such as pharmaceuticals, where precise transformations are required. Traditional methods rely on time-consuming experimental screening, but recent advances in machine learning and databases offer a more targeted approach. This project explores using computational tools to identify and generate enzymes from the vast and diverse world of bacterial secondary metabolism, aiming to streamline the selection process for complex biocatalytic reactions and enhance the development of greener, more efficient industrial processes.

Job Advertisement

Mental illnesses, such as depression, are likely to be heterogeneous constructs where several different neurobiological impairments are likely to be diagnosed as one illness. Concretely, this means that patients with the same diagnosis may have very different causes, which makes it unlikely that any intervention would work for all of them alike. The goal is thus to establish novel, neurobiologically-informed assessments that can detect and target specific (impaired) brain processes. In this project, we will target the reward system, a candidate system for mood and motivation impairments. We will use existing and new behavioural datasets and artificial intelligence to establish novel reward-related profiles. These profiles may be used to improve assessments in mental health and move towards a personalised psychiatry.

Application closed

Medical and pharmaceutical innovation relies on clinical trials as a crucial quality control step. Most trials are delayed or altogether fail due to unrealistic trials design. This project aims to fundamentally change the way in which clinical trials are designed and recruited for, by using foundation models to get better insights into eligible cohorts, recruitment bottlenecks and efficiently discover lessons learned during thousands of historic clinical trials. It lives at the intersection of basic scientific innovation of NLP models and applied impact on the life sciences.

Application closed

Research Interests

• biology with a focus on cell culture, molecular biology and transcriptional regulation
• molecular medicine in diabetes prevention
• data science and bioinformatics

Project: Multi-modal deep learning for biomarker discovery in mass spectrometry imaging data

Tissue microenvironment and the spatial organization of cells in the tissue are strongly associated with physiological and pathological processes. Mass spectrometry imaging (MSI) data contains distribution patterns of numerous molecules and their molecular masses. This research project will develop novel machine learning methods to associate MSI data with preclinical phenotypes, with the goal to discover novel molecular markers (as toxicity or efficacy markers or to highlight unexpected metabolic reactions).

Christoph's LinkedIn Profile

Research Interests

• theoretical foundations of machine learning, including:
  • out-of-distribution detection
  • uncertainty quantification
  • Bayesian machine learning
  • physics-informed neural networks
• analog-to-digital conversion algorithms
• applied mathematics

Project: Anomaly detection for histopathology to optimize drug discovery

We conduct research on anomaly detection for histopathology. The goal is to find any deviation from healthy tissue, such as pathologies, based on image data. The approach should learn the appearance of healthy tissue, be able to take advantage of certain known pathologies, but in particular detect completely new changes. Furthermore, it should be able to improve continuously getting different kind of human feedback from pathologists. Fast detection of anomalous tissue is important for early detection of toxic drug candidates.

Olga's Google Scholar Profile

Contact Olga

Research Interests

• Cutting-edge Science and innovative Technologies
• Single-cell and bulk Transcriptomics
• Machine Learning to improve biomedical research
• Human Health & Pharma

Project: Transcriptomics machine learning models

This project will build hybrid transcriptomics machine learning models for drug discovery. Unsupervised learning will be used to build cell-type specific and integrated bulk tissue modules, then supervised learning will be applied to connect those modules to a holistic hybrid machine learning model for transcriptomics.

Contact Maik

Research Interests

• Medical Informatics
• Quantitative Morphology
• Genomics and Transcriptomics
• Artificial Intelligence for Healthcare
• Representation Learning

Project: Learning joint representations of EHR and omics data

The project will work on large biobank data combining EHR and genetic information to learn the best joint representation that improves downstream analyses. Those downstream analyses include disease subtyping, disease risk prediction, and patient stratification.

Contact Tobias

Research Interests

• Information Retrieval

• Natural Language Processing

• Generative AI

Project: Language Model-guided Cohort Discovery

Medical and pharmaceutical innovation relies on clinical trials as a crucial quality control step. Most trials are delayed or altogether fail due to unrealistic trials design. This project aims to fundamentally change the way in which clinical trials are designed and recruited for, by using foundation models to get better insights into eligible cohorts, recruitment bottlenecks and efficiently discover lessons learned during thousands of historic clinical trials. It lives at the intersection of basic scientific innovation of NLP models and applied impact on the life sciences.

Shrestha’s Website

Email Shrestha

Research Interests

• probabilistic deep modelling
• modelling of dynamical systems and time series 
• medical applications
• model discovery

Project: Simulation-based model discovery for time-series models in quantitative systems pharmacology

This interdisciplinary project aims to develop probabilistic machine learning tools for automatically inferring mechanistic dynamical system models from time series observations. The machine learning tools will be adapted to study disease mechanisms that underly Chronic Kidney Disease by extending existing limited models.

Email Daniel