Date:
27 June 2025

Title
Artificial intelligence in mental health: novel routes towards personalized medicine

Abstract
Artificial intelligence (AI) offers a promising framework for advancing personalized medicine applications in psychiatry. This talk will focus on the development of two AI-driven pipelines with application to genetics, as well as neuroimaging data. The first leverages AI to integrate genetic risk contributions to a large number of mental health relevant phenotypes in genetic stratification models with application to schizophrenia. The second presents a deep learning-driven procedure to perform advanced normative modeling on structural magnetic resonance imaging data with application to alcohol use disorder and depression. These approaches provide deeper insight into the genetic and neurobiological architecture of mental disorders and may inform future, personalized approaches towards clinical management, therapeutic development and prevention. 

Date
25 April 2025

Title
Hybrid AI and Open Source for Drug Design

Abstract
Addressing the complexities of drug design – exemplified here on human protein kinases – necessitates innovative approaches that blend artificial intelligence (AI) with domain expertise. With over 6,000 human kinase structures available in the PDB and around 70 small molecule kinase inhibitors, persistent challenges such as drug promiscuity, resistance, and unexplored kinase territories remain.

This presentation showcases open-source approaches that integrate domain knowledge into AI frameworks to overcome data scarcity issues and enhance model generalization. Leveraging openly available kinase data, we demonstrate how hybrid AI and classical methods can generate new insights and foster community engagement. The TeachOpenCADD platform [1,2] serves as a versatile tool for orchestrating diverse computer-aided drug design (CADD) tasks, exemplified on individual kinases. We introduce freely available resources to support kinase research and incorporate structural data to enhance prediction accuracy and guide drug discovery efforts. These resources include active learning on the KinFragLib kinase fragment library for inhibitor design [3], and structure-based deep learning approaches for affinity prediction [4].

These projects highlight how hybrid AI approaches fuse experimental with computational insights to advance the frontier of data-driven drug design.

References:
[1] S. Dominique, et al., TeachOpenCADD 2022: Open Source and FAIR Python Pipelines to Assist in Structural Bioinformatics and Cheminformatics Research. Nucleic Acids Research, 2022. https://doi.org/10.1093/nar/gkac267
[2] M. Backenköhler, et al., TeachOpenCADD goes Deep Learning: Open-source Teaching Platform Exploring Molecular DL Applications. ChemRxiv, 2023. https://doi.org/10.26434/chemrxiv-2023-kz1pb
[3] S. Dominique, et al., KinFragLib: Exploring the Kinase Inhibitor Space Using Subpocket-Focused Fragmentation and Recombination. Journal of Chemical Information and Modeling, 2020. https://doi.org/10.1021/acs.jcim.0c00839
[4] M. Backenköhler, J. Groß, V. Wolf, A. Volkamer, Guided docking as a data generation approach facilitates structure-based machine learning on kinases. Journal of Chemical Information and Modeling, 2024, 64, 10, 4009–4020  https://pubs.acs.org/doi/10.1021/acs.jcim.4c00055

Vita:
Andrea Volkamer studied bioinformatics at Saarland University, followed by a one-year research stay at Purdue University (USA). After her PhD at the University of Hamburg, she worked at the BioMedX Innovation Center Heidelberg as a Postdoc on methods for the development of selective kinase inhibitors. Before her appointment as a university professor at Saarland University in 2022, she has been leading a research group for structural bioinformatics and in silico toxicology at the Institute of Physiology at Charité Universitätsmedizin Berlin as an assistant professor. Andrea Volkamer is an associated researcher at the Helmholtz Institute for Pharmaceutical Research Saarland (HIPS).

Andrea Volkamer’s scientific focus is on method development with a particular interest in structure-based machine learning approaches for computational drug design. The application focus so far is in the areas of cancer research and risk assessment of new molecules (alternative to animal testing). She does so by exploiting the rapidly growing amount of available experimental data – increasingly complemented by synthetic data, as well as powerful computational algorithms – to develop in silico methods that support the drug design process. 

Andrea Volkamer is a strong advocate of FAIR science and supports many open source initiatives. In this context, for example, her public teaching materials TeachOpenCADD and Dr. med. AI are worth mentioning, as well as her contributions to the Covid Moonshot Consortium.

(Based on: https://saarland-informatics-campus.de/en/piece-of-news/andrea-volkamer-new-professor-for-data-driven-drug-development/ )

Welcome - Volkamer Lab

Date:
8 November 2024

Title:
Federated and privacy-preserving analytics and machine learning for drug discovery and precision medicine

Abstract:
Federated analytics (“bring the computation to the data”) is a key emerging trend in the analysis of sensitive biomedical data. It implements – at the level of statistical and machine learning models – the central data minimization requirement of the GDPR by limiting the exchange of information between data controllers (and one or more aggregation hubs) to aggregate statistical data. If further protection of patient privacy or data confidentiality is necessary, such schemes can be turned into privacy-preserving algorithms that combine statistical aggregation with privacy-preserving analytics techniques (homomorphic encryption and multiparty computation). The increased protection comes nevertheless at the cost of increased computational load/latency and decreased flexibility. In this talk, I will address applications of these principles in several proof-of-concepts in genetic association studies, variant analysis, pharmaceutical drug-target activity prediction. I will also discuss how such strategies help address key requirements of the GDPR.

Vita:
Yves Moreau is a professor at the University of Leuven, Belgium. His team focuses on AI algorithms and software platforms for the integration of complex data in clinical genomics and drug discovery: (1) federated analysis of real-world clinical and genomic data, (2) data fusion algorithms for the identification of pathogenic genetic variation in rare genetic disorders and liquid biopsies, and (3) data fusion for drug discovery and drug design. At the algorithmic level, he focuses on the development of novel AI methods, such as deep learning and Bayesian matrix factorization, for the fusion of heterogeneous sparsely-observed data; and on privacy-preserving implementations of such methods. He aims at demonstrated clinical or industrial applicability and proven effectiveness in human genetics research and drug discovery.

He is also engaged in a reflection on how information technology and artificial intelligence are transforming our world and on how to make sure this transformation is beneficial for all. In particular, he is actively pushing back against the emergence of surveillance societies that has been made possible by such technological advances.

Date:
29 July 2024

Title:
AI and the future of radiology

Abstract:
Artificial Intelligence (AI) is changing many fields across science and our society. This talk will discuss how AI is changing medicine and healthcare, particularly in radiology. I will focus on how AI can support the acquisition of medical images and image analysis and interpretation. This can enable the early detection of diseases and support the improved personalised diagnosis. I will show several examples of this in the talk, including neuro and cardiovascular MR imaging. Furthermore, we will discuss how AI solutions can be privacy-preserving while also providing trustworthy and explainable solutions for clinicians.

Vita:
Since 2020, Daniel Rückert is Alexander von Humboldt Professor for AI in Medicine and Healthcare at the Technical University of Munich. He is also a Professor at Imperial College London. He gained a MSc from Technical University Berlin in 1993, a PhD from Imperial College in 1997, followed by a post-doc at King’s College London. In 1999 he joined Imperial College as a Lecturer, becoming Senior Lecturer in 2003 and full Professor in 2005. From 2016 to 2020 he served as Head of the Department of Computing at Imperial College.

Source: https://www.professoren.tum.de/en/rueckert-daniel

Title:
The origin of genetic coding: how did computation subjugate autonomous molecular processes?

Abstract:
Symbolic information processing, computation, first emerged locally in the universe when life originated on the surface of our planet. This event required the coupling of information copying/storage (nucleic acid replication) and information transformation/interpretation (coded protein synthesis, "translation"). The initial coupling of these two fundamental molecular biological processes is problematic in two ways. First, all ordinary chemical explanations of genetic replication-translation coupling fail the test of thermodynamic possibility. And second, computational explanations have to overcome the chicken-egg bootstrapping problem of interpreting genetic software to create enzymic hardware that is required for the programmed interpretation of genetic information. I will discuss the shift in scientific thinking about the matter-information nexus that is required to understand how these theoretical problems can be solved; and I will give a personal perspective on how such a change in thinking bears on humanity's global problems concerning the balancing of natural constraints and technological freedom: climate change, bioengineering, artificial intelligence.

Vita:
Peter Wills is from the Department of Physics at the University of Auckland, New Zealand. His initial work on the self-organisation of genetic coding was conducted in Manfred Eigen's group in Göttingen during a sabbatical stay in 1989. His more recent work concerns biochemical, computational and mathematical analyses of the evolutionary emergence of the enzymes responsible for genetic coding, the aminoacyl-tRNA synthetase (AARS) catalysts. He also has enduring interests in the fields of statistical thermodynamics and prion replication. He is known publicly for his political activism on issues in which science plays a crucial role, especially nuclear weapons and the natural environment.

Title:
Immune escape and adaptation of human RNA viruses

Abstract:
Over the past 4 years, SARS-CoV-2 has repeatedly given rise to new more transmissible or immune evasive variants. This adaptive evolution of the virus has profoundly shaped how the pandemic unfolded and was followed in near real-time through large scale global genomic surveillance. I will give an overview of SARS-CoV-2 evolution, compare it to the evolution of influenza viruses, and discuss how we can use the now abundant sequencing data to learn about the potential for future viral adaptations.

Vita:
https://www.biozentrum.unibas.ch/de/forschung/research-groups/research-groups-a-z/own-content/unit/forschungsgruppe-richard-neher/neher-cv

Title:
Deciphering the chemical language of the microbiome

Abstract:
Microbial specialized metabolites are important mediators of molecular interactions between microbes as well as with their plant, animal, or human host. In a way, they constitute the ‘chemical language’ of the microbiome. Interpreting this language is key to understanding the mechanistic basis for many microbiome-associated phenotypes, such as disease suppression or growth promotion.Genome sequence data has revealed that only a tiny fraction of the chemical diversity of these natural products has been unearthed. In recent years, a range of computational methods have been developed to identify these molecules and the metabolic gene clusters that encode their production, and to assess their biological activities. Here, I will highlight recent work performed in my research group on developing and applying these approaches to accelerate natural product discovery, as well as to study the roles of these pathways in microbe-microbe and host-microbe interactions in microbiomes. Specifically, I will provide examples of how we are applying these methods to predict functions of unknown biosynthetic gene clusters and to connect them to microbiome-associated phenotypes.

Vita:
Marnix Medema is a Professor of Bioinformatics at Wageningen University. His research group develops and applies algorithms for the (meta)genomic identification and functional prediction of microbial biosynthetic pathways, with the aim to unravel the chemical language of microbiomes. He built and co-coordinates the development of the antiSMASH software for identification of biosynthetic gene clusters and developed various additional algorithms to chart their diversity and identify their functional roles in microbiomes. Medema is recipient of NWO Rubicon, Veni and Vidi fellowships and an ERC Starting Grant, and has coordinated several international consortia studying bacterial specialized metabolites. He received several prizes for his work, including the NBIC Young Investigator Award. He is editorial board member of Natural Product Reports, mSystems and FEMS Microbes, senior editor of ISME Communications. Also, he is member of the scientific advisory board of Hexagon Bio and co-founder of Design Pharmaceuticals. From 2020-2022, he also serves as Van der Klaauw visiting professor of theoretical biology at Leiden University.

Title:
The Value of Open Data. Innovative Bioinformatics Projects and Partnerships by EMBL-EBI: AlphaFold, Open Targets, and other Examples.

Abstract:
EMBL's European Bioinformatics Institute (EMBL-EBI) maintains the world’s most comprehensive range of freely available and up-to-date molecular data resources. Developed in collaboration with our colleagues worldwide, our services promote to share data, perform complex queries and analyse the results in different ways.In my presentation I like to highlight how our open data resources enable AI approaches and give some recent examples of innovative EMBL-EBI AI projects and partnerships. One of these is AlphaFold, a state-of-the-art AI system developed by DeepMind, which is able to computationally predict protein structures with unprecedented accuracy. These predictions are being made freely and openly available to the global scientific community in partnership with EMBL-EBI. Open Targets is an innovative, large-scale, multi-year, public-private partnership that uses human genetics and genomics data for systematic drug target identification and prioritisation.These and related projects are opening up new and exciting research avenues to dramatically deepen our understanding of human health, disease and our environment, with implications for areas like drug design and sustainability.

Vita:
Rolf Apweiler is Director of EMBL-EBI, together with Ewan Birney. Prior to this position he was Joint Associate Director, after many years of leading protein resources such as UniProt and InterPro. Rolf has made a major contribution to methods for the automatic annotation of proteins, making it possible to add relevant information to proteome sets for entire organisms. He has spearheaded the development of standards for proteomics data, and his teams have maintained major collections of protein identifications from proteomics experiments (PRIDE) and molecular interactions (IntAct). He also led EMBL-EBI’s contribution to the Gene Ontology, was Director of Open Targets, and is now leading the efforts of EMBL-EBI around the European COVID-19 Data Platform.

Rolf received his PhD from the University of Heidelberg in 1994, and has been at EMBL since 1987. His major contribution to the field of proteomics was recognised by the the Human Proteomics Organisation’s “Distinguished Achievement Award in Proteomics” in 2004 and his election to President of the Human Proteomics Organisation, which he held in 2007 and 2008. In 2012, he was elected as a member of EMBO and in 2015 he was elected to an ISCB (International Society for Computational Biology) fellow. Rolf also served over many years on a multitude of Editorial Boards and Scientific Advisory Boards.

Source: EMBL-EBI

Title:
Multimodal Machine Learning in Medicine

Abstract:
In this talk, I will touch upon some of the challenges and chances that arise in the area of machine learning in medicine. I will put special emphasis on dealing with the multiple heterogeneous data types that naturally co-occur in medical practice. I will present different types of generative models for multimodal learning and demonstrate the need of multimodal modals in medicine on several medical application examples.

Vita:
Julia Vogt is an assistant professor in Computer Science at ETH Zurich, where she leads the Medical Data Science Group. The focus of her research is on linking computer science with medicine, with the ultimate aim of personalized patient treatment. She has studied mathematics both in Konstanz and in Sydney and earned her Ph.D. in computer science at the University of Basel. She was a postdoctoral research fellow at the Memorial Sloan-Kettering Cancer Center in NYC and with the Bioinformatics and Information Mining group at the University of Konstanz. In 2018, she joined the University of Basel as an assistant professor. In May 2019, she and her lab moved to Zurich where she joined the Computer Science Department of ETH Zurich.

Source: ETH Zurich

Title:
The CoVerage genomic surveillance platform predicts and characterizes SARS-CoV-2 Variants of Interest

Abstract:
Rapidly evolving viral pathogens such as SARS-CoV-2 continuously accumulate amino acid changes, some of which affect transmissibility, virulence or improve the virus’ ability to escape host immunity. Since the beginning of the pandemic and establishment of SARS-CoV-2 as a human pathogen, multiple lineages with concerning phenotypic alterations, so called Variants of Concern (VOCs), have emerged and risen to predominance. To optimize public health management and to ensure the continued efficacy of vaccines, the early detection of such variants of interest is essential. Therefore, large-scale viral genomic surveillance programs have been initiated worldwide, with data being deposited in public repositories in a timely manner. However, technologies for their continuous interpretation are currently lacking. Here, we describe the CoVerage web platform for viral genomic surveillance, which continuously predicts and characterizes novel and emerging potential Variants of Interest (pVOIs) together with their antigenic and evolutionary alterations. Using the establishment of Omicron and its current sublineages as an example, we demonstrate how CoVerage can be used to quickly identify and characterize such variants. CoVerage can facilitate the timely identification and assessment of future SARS-CoV-2 Variants of Concern.

Vita:
Alice Carolyn McHardy holds a diploma in biochemistry and a doctoral degree (Dr. rer. nat) in bioinformatics, both from Bielefeld University in Germany. From 2005 to 2007 she first was a postdoc and then a permanent staff member in the Bioinformatics and Pattern Discovery Group at the IBM T.J. Watson Research Center in Yorktown Heights, USA.

She then became the head of the independent research group for Computational Genomics and Epidemiology at the Max Planck Institute of Computer Science in Saarbrücken. In 2010, she was appointed Chair of Algorithmic Bioinformatics at Heinrich Heine University in Düsseldorf.

In 2014, she became head of the Department of Computational Biology for Infection Research at the Helmholtz Centre for Infection Research in Braunschweig and was appointed as a full professor at TU Braunschweig.

Source: Helmholtz Centre for Infection Research

The MicrobeAtlas: A Global Survey of Microbial Diversity.
(14 July 2023, M3 Building)

Spatial Single-Cell Data Visualization with the Vitessce Framework.
(30 June 2023 at 1 p.m.)

A few reflections on doing science.
(10 February 2023 at 2 p.m.)

Learning the temporal dynamics of gene expression from single-cell transcriptomics data.
(9 December 2022 at 2 p.m.)

Data Anonymisation in Theory and Practice.
(28 October 2022 at 2 p.m.)

The Shock of the New - Machine Learning reveals the true complexity of the proteome.
(22 April 2022 at 2 p.m.)

Making uncertainty explicit when analyzing microbiomes.
(18 March 2022 at 2 p.m.)