Location: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Computational models and analyses play a crucial role in drug research and development across various stages. These models have become an integral part of the drug development process, addressing a range of challenges. There are two main types of approaches used in these applications: 1) Mechanistic models are utilized to address smaller scale biology and engineering related questions. They help optimize drug design, target engagement, dosing, and PK/PD (pharmacokinetics/pharmacodynamics). 2) Machine learning and statistical models are employed for larger-scale OMICS data analysis. They aid in identifying and characterizing targets, selecting patient populations, and understanding biomarkers of response or resistance. This presentation will showcase recent examples in both areas and explore potential future opportunities for Systems Biology modeling in drug research and development.

Speaker: Dr. Andreas Raue

Biography

Dr. Andreas Raue's research focuses on exploring and developing novel therapeutic approaches in oncology by harnessing advanced technologies such as single-cell sequencing and deep learning. He earned his Ph.D. from Albert-Ludwigs University in Freiburg, Germany, before relocating to the United States. Over the span of more than a decade, he contributed significantly to the fields of biotechnology and pharmaceuticals, conducting discovery and translational research at Merrimack Pharmaceuticals, HiFiBiO Therapeutics, and Novartis. Throughout his tenure, Dr. Raue and his team successfully advanced numerous projects from initial target identification to clinical trials. Until recently, Dr. Raue led the Immuno-Oncology and Hematology Data Science team at Novartis, spearheading efforts to develop and deliver state-of-the-art therapies aimed at improving outcomes for patients with cancer and blood disorders. From May 2024, he will hold the Chair of Modelling and Simulation of Biological Processes at the University of Augsburg.

Location: Large lecture hall (2nd floor, room 047, University Hospital)

Abstract

Clinical decision-making is a complex process that requires the integration of a wide range of patient information and extensive medical knowledge. To effectively support this process, clinical decision support systems should not only be able to form a holistic view of the patient and consider prior knowledge but also be able to convey their reasoning. In light of recent advances in large language models (LLMs), this talk explores the potential of multimodal foundation models in enabling such systems. We first discuss the clinical reasoning process and then present how the integration of knowledge and multimodal patient representations in LLMs could interactively support clinicians in their decision-making. Finally, recent works in radiology report generation and conversational assistance are showcased to illustrate the potential impact of these technologies on radiology and other medical fields.

Speaker: Dipl.-Ing. Matthias Keicher (Research Manager at the Chair of Computer Aided Medical Procedures, Technical University of Munich)

Biography

Matthias Keicher is a senior PhD candidate and group leader at the Chair for Computer Aided Medical Procedures at the Technical University of Munich (TUM) and the interdisciplinary research lab of the hospital Klinikum rechts der Isar in Munich. He leads a research group focused on applying deep learning for vision and language understanding in radiology and neuroradiology, focusing on topics such as automated report generation, structured reporting, and visual question answering. His research interests revolve around using multimodal deep learning to enable holistic clinical decision support systems that can reason over comprehensive patient data. He graduated as a mechanical engineer at TUM specializing in biomedical engineering. Before returning to academia, he gained several years of industrial experience in the healthcare domain, working as a product manager, director of business development, and eventually as CTO and managing director. He also co-founded a health technology startup during this time.

Location: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

We will introduce the concept of computational modeling & simulation of organ systems in the human body using the example of cardiac electrophysiology. According to the multi-scale nature of the biological system, we will start from the ion channel level and work our way up to the cellular and eventually organ scale.

We will then use the concept of ensemble simulation spaces to discuss how modeling and simulation can be employed for digital twin applications enabling personalized medicine and for creating large virtual cohorts of digital chimeras enabling in silico trials and machine learning solutions.

Applications include the identification of the best ablation and/or pharmacological strategy to treat atrial fibrillation and prediction of therapy success.

Speaker: PD Dr.-Ing. Axel Loewe (Head of Computational Cardiac Modeling, Karlsruhe Institute of Technology)

Biography

Dr. Axel Loewe heads the Computational Cardiac Modeling Group at Karlsruhe Institute of Technology with a focus on cardiac electrophysiology and biomechanics. His group is committed to method development and the application of computational models to answer questions of clinical relevance at the intersection of engineering, computer science, and medicine. To develop methods and conduct simulation studies, methods of software engineering, algorithmics, numerics, signal processing, data analysis, and machine learning are used. Axel has a track record of fruitful collaboration with leading clinicians to optimize diagnosis and therapy of cardiac diseases.

He studied Electrical Engineering and Information Technology in Karlsruhe and Stockholm and earned a PhD with distinction in Biomedical Engineering in 2013 from KIT. Habilitation "Modeling and Simulation for Medicine" in 2021. Reviewer and guest editor for >40 journals in the fields of Biomedical Engineering, Cardiology and Machine Learning >30 academic distinctions including Patient Safety Award 2018, Gips-Schüle Preis 2018, CinC Young Investigator award 2016.

Location: Small lecture hall (2nd floor, room 048, University Hospital)

Abstract

Our world is 3D and so is the patient. But visual camera observations are 2D. Advancements in digital imaging now enable capturing rich 3D information of our surrounding, transforming our ability to digitally perceive, present, and analyze data. By combining multiple data sources or frames in a video, we can create context-enhanced digital copies of the physical world as well as the patient - and apply data-driven interpretations and processing. Data sources reach from ceiling mounted cameras in the OR to endoscopic images or robot-mounted sensors.
In this talk, we want to look into underlying core ideas of example 3D computer vision pipelines and investage how these approaches can be used in clinical applications.
As a basis for a discussion at this exciting interdisciplinary crossroad, we dive into the subtopics 3D digital reconstruction, data curation for XR, and robot-assisted surgery.

Speaker: Dr. Benjamin Busam (Chair of Computer Aided Medical Procedures, Technical University of Munich)

Biography

Benjamin Busam is a Senior Research Scientist with the Technical University of Munich. He coordinates the Computer Vision activities at the Chair for Computer Aided Medical Procedures, I16. Formerly Head of Research at FRAMOS Imaging Systems, he led the 3D Computer Vision & AI Team at Huawei Research, London from 2018 to 2020. Benjamin studied Mathematics at TUM (Germany), ParisTech (France) and at the University of Melbourne (Australia), before he graduated with distinction at TU Munich in 2014. In continuation to a mathematical focus on projective geometry and 3D point cloud matching, he now works on 2D/3D computer vision and sensor fusion and applications into the medical domain. He For his work on adaptable high-resolution real-time stereo tracking he received the EMVA Young Professional Award 2015 from the European Machine Vision Association and was awarded Innovation Pioneer of the Year 2019 by Noah's Ark Laboratory, London. He was given multiple Outstanding Reviewer Awards at 3DV 2020, 3DV 2021, and ECCV 2022.

Location: Small lecture hall (2nd floor, room 048, University Hospital)

Abstract

How can positron emission tomography and modelling techniques be used to quantify receptor availability in the brain? Positron emission tomography (PET) is an imaging technique that can be used to visualise metabolic processes. Small amounts of a radioactively labelled molecule that is involved in a metabolic pathway of interest are administered into the body. Radioligands that bind specifically to neuroreceptors are used for neurological or psychiatric issues. Compartment models can be used to obtain quantitative physiological parameters. Usually, a complex imaging protocol is required, which, in addition to the collection of tissue data by PET, also requires arterial blood data as input function for kinetic models. Using the endocannabinoid system as an example, it is shown how PET and modelling methods can be used to address translational and hypothesis-driven questions.

Speaker: PD Dr. Isabelle Miederer (Clinic and Polyclinic for Nuclear Medicine, University Medical Center Mainz)

Biography

The imaging of neuroreceptor systems, in particular the endocannabinoid system, is the research focus of Dr Isabelle Miederer. She has made an important contribution to PET methods for quantifying receptor availability and glucose metabolism in the brain in both preclinical and clinical research. In the field of data processing, she mainly applies the methods of tracer kinetic modelling. In a translational project funded by the German Research Foundation, she is currently investigating the role of the endocannabinoid system in attention-deficit/hyperactivity disorder (ADHD).

Dr Isabelle Miederer studied Biomedical Engineering at Furtwangen University and received her doctorate from the Faculty of Computer Science at the Technical University of Munich. She habilitated in Experimental Nuclear Medicine at Johannes Gutenberg University Mainz on the subject of ‘Methods for PET imaging of the endocannabinoid system in preclinical research’.

Location: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

In today's healthcare landscape, the ability to transform vast amounts of raw clinical data into actionable knowledge is a crucial step toward improving patient outcomes and advancing medical research. Semantic data integration plays a key role in bridging and unlocking data silos, enhancing data interoperability, and uncovering hidden insights. Through real-world examples and results from our junior research group, IMPETUS, this talk will provide insights into the practical applications of semantic data integration within the UKSH MeDIC.

Speaker: Dr. Hannes Ulrich (Institute of Medical Informatics and Statistics, Christian-Albrechts-University Kiel)

Biography

Dr. Hannes Ulrich is a postdoctoral researcher at the Institute for Medical Informatics and Statistics, section of Medical Informatics, University of Kiel, and University Hospital Schleswig-Holstein, where he leads data acquisition team of the Medical Data Integration (MeDIC) and serves as deputy leader of the IMPETUS junior research group. His research focuses on metadata-driven data integration within clinical settings to enable routine data for the secondary use, improve patient outcomes and advance medical research. He completed his doctorate in Medical Informatics at the University of Lübeck, with an award-winning dissertation on standardized metadata integration.

Location: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Specific phobia is a severe, widespread and irrational fear that may be associated with objects (e.g. spiders) or situations (e.g. enclosed spaces) that pose little or no danger. Studies of the neural basis of specific phobia often focus on a predicted fear network, leading to a pre-selection of brain regions to be studied. This pre-selection carries the risk of false-negative results, as large parts of the brain are not considered. Using the example of individuals diagnosed with spider phobia and a parallel control group who underwent functional magnetic resonance imaging in the task-independent resting state, the talk will present a data-driven method that overcomes these limitations. Multivariate pattern analysis, which considers the totality of functional connections in the brain (the connectome), was used to identify regions of maximum variance between experimental groups. The results were further characterised for behavioural relevance using metadata and post-hoc analyses.

Speaker: Prof. Dr. Markus Mühlhan (Chair of Neuroscience, Medical School Hamburg)

Biography

Markus Mühlhan is Professor of Neuroscience at the Medical School Hamburg where he investigates neuroendocrinological and neurocognitive changes in stress, anxiety and substance use disorders at the Institute of Cognitive and Affective Neurosciences (ICAN). The methodological focus is on magnetic resonance imaging and neuroimaging meta-analyses as well as the behavioural characterisation of MRI findings using metadata. Markus initially trained as an electronics technician before studying psychology at the University of Oldenburg. He received his PHD at the Faculty of Mathematics and Natural Sciences of the TU-Dresden and worked at the Institutes for Biopsychology and Clinical Psychology.  Since 2019 Markus is full Professor of Neurosciences at the Medical School Hamburg.

Location: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Rare genetic variants can strongly predispose to disease, yet accounting for rare variants is statistically challenging, and principled strategies for integrating possibly diverse types of variant annotations in a data-driven manner are lacking. Here, we present DeepRVAT (Deep Rare Variant Association Testing), a deep set model that learns gene impairment scores from rare variants, annotations, and phenotypes. DeepRVAT infers the relevance of different annotations and their combination directly from data, eliminating ad hoc modeling choices that characterize existing methods. DeepRVAT estimates a single, trait-agnostic gene impairment score for each gene in each sample, enabling both risk prediction and gene discovery in a unified framework and seamless integration into established association testing frameworks. We apply DeepRVAT on 34 quantitative and 63 binary traits across 469,382 whole-exome-sequenced individuals from the UK Biobank. We integrate state-of-the-art annotations, including AlphaMissense, PrimateAI, AbSplice, DeepRipe, and DeepSEA, and find a substantial increase in gene discoveries and improved replication rates on held-out individuals over previous methods.  We demonstrate the applicability of pre-trained DeepRVAT models to new traits, facilitating the study of disease cohorts with limited training data. Furthermore, we significantly improve the detection of individuals at high genetic risk by combining common variant polygenic risk scores with DeepRVAT.

Speaker: Eva Holtkamp, M.Sc. (PhD student in the Chair of Computational Molecular Medicine, TUM)

Biography

Eva Holtkamp is a PhD student in computational biology in the group of Prof. Julien Gagneur at the Technical University of Munich and the Helmholtz Center Munich. She is part of the Munich Data Science School PhD program. Previously, she earned a master’s degree in molecular biotechnology with a focus on bioinformatics from Heidelberg University. Her research focuses on using deep learning and statistical methods to understand the effects of rare genetic variants on common polygenic traits and diseases by leveraging genome sequencing data from biobank-scale cohorts.

Location: Small lecture hall (2nd floor, room 048, University Hospital)

Abstract

Clinical research and routine care data are often separated and poorly integrated. In this talk, we will present data integration and management solutions for feasibility studies, recruitment of potential study participants, collection of patient-reported outcomes, and clinical study documentation developed and used at the Münster University Hospital.

Speaker: Dr. Michael Storck (Group Leader Medical Data Integration Centre, University of Münster)

Biography

Dr. Michael Storck is the head of the Medical Data Integration Centre ( MeDIC) at the Institute of Medical Informatics in Münster. He is spokesperson of the Working Group "Medical Software and Medical Device Regulation" of the TMF e.V.. He studied computer science at the University of Münster, earned his doctorate in medical informatics and conducts research in the field of interoperability and secondary use of clinical routine data.

Location: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

In this talk, I will review some recent developments in the field of interpretable AI models for genomics data.

These models can be used to predict perturbation effects from drug treatment of genetic effects, or infer gene regulatory networks.

Speaker: Prof. Dr. Carl Herrmann (Professor Dept. Bioinformatics, Institute of Pharmacy and Molecular Biotechnology, University Heidelberg)

Biography

Prof. Dr. Carl Herrmann has been a Professor at the Department of Bioinformatics, Institute of Pharmacy and Molecular Biotechnology at the University of Heidelberg since 2023. From 2018 to 2023, he was an Assistant Professor at the Medical Faculty of the University of Heidelberg. Between 2013 and 2018, he led a research group in the Division of Theoretical Bioinformatics at DKFZ Heidelberg and was an Assistant Professor at the IPMB, University of Heidelberg. From 2003 to 2013, he served as an Assistant Professor at the University of Marseille and IBDML, CNRS (France).

Prof. Herrmann has also been involved in various research and training initiatives, including serving on the Board of Directors of the Research and Training Group "Big Data Research in the Biosciences" from 2018 to 2021, and as a member of the HGS MathComp Graduate School since 2021.