Our group focuses on the design, implementation and evaluation of scalable tools for genomic sequence analysis (Bioinformatics) and computational science applications. Our approach is based on using modern high performance computing (HPC) technologies (such as CUDA-enabled GPUs and GPU clusters) in order to design efficient parallel algorithms that serve as a foundation for a wide variety of tools such as multiple sequence aligners, de-novo genome assemblers, short-read aligners, short-read clustering and sequence database searching tools. Our methods and tools are often developed in collaboration with interdisciplinary partners at JGU Mainz, such as the Department of Biology, the Mainz Medical School, and the Institute of Molecular Biology.
For the pioneering work in the area of CUDA-enabled bioinformatics, we have been awarded the status of a GPU Research Center and a GPU Education Center.
Data compression, Vector-Matrix-Multiplication, General-purposed Sparse Matrix-Matrix Multiplication, and more are important Algorithms in parallel Applications.
Other fields such as atmospheric physics, AI, and more also benefit from fast and efficient computing.
Retrieving comprehensible rule-based knowledge from medical data by machine learning is a beneficial task, e.g., for automating the process of creating a decision support system. While this has recently been studied by means of exception-tolerant hierarchical knowledge bases (i.e., knowledge bases, where rule-based knowledge is represented on several levels of abstraction), privacy concerns have not been addressed extensively in this context yet. However, privacy plays an important role, especially for medical applications.
Introduction to Studying Privacy Aspects of Learned Knowledge Bases in the Context of Synthetic and Medical Data
How decentralized should data be stored to protect privacy? To what extent does this affect the transparency of data and algorithms? These conflicting goals will be analyzed in a research center for machine learning.
The goal of the project is to establish an interdisciplinary research center for Machine Learning at the University of Mainz. Here, interactions and dependencies of different properties of Machine Learning will be studied. The research topics which will be investigated include transparency and fairness of data and algorithms, as well as data protection requirements and efficient use of resources such as electricity. A strong focus will be given to their competing needs. As an example, how decentralized can data be stored and processed to protect privacy? To what extent does decentralization affect the transparency of algorithms and data? What impact does this have on energy consumption? These various trade-offs will be identified and characterized to create workable trade-offs. Ethical and legal aspects will be considered.
The research project is scheduled for 6 years and is funded by the Carl Zeiss Foundation.
As part of the project, a tenure-track endowed professorship in the field of trustworthy AI will be established.
An important element in the project is communication within the Johannes Gutenberg University, but also with the University of Applied Sciences Mainz, with external research institutions and the regional economy. To foster collaboration, regular workshops will be held towards the end of each project year. The workshops will focus on advancing the development and implementation of trustworthy AI methods in Mainz.
Transregio 319 RMaP is a joint venture combining:
Molecular Biology, Biochemistry, Cellular Biology, Chemical Biology, Biophysical Chemistry, Structural Biology, Developmental Biology, Genetics, Bioinformatics.
Epitranscriptomics rely heavily on modification mapping techniques, requiring fast and effective treatment of massive amounts of sequencing data. New modification mapping chemistries lead to new data formats and, like conventional ones, require considerable efforts for data treatment and analysis. Here, a new chemistry will be developed along with mapping, and base calling techniques in a harmonized approach. Standards and tools for quality and performance assessment will be developed to improve accuracy and speed, including machine learning, ultra-fast implementations, and easy-to-use pipelines, which will be applicable to other mapping chemistries, as well.
Deutsche Forschungsgemeinschaft Subproject C01 Novel modification mapping techniques and efficient RNA-Seq data management
Machine learning in big data of atmospheric physics
Big Data in Atmospheric Physics (BINARY) is an interdisciplinary project, involving the research fields Atmospheric Physics and Computer Sciences. Researcher from the Institutes of Atmospheric Physics and Computer Sciences at the Johannes Gutenberg University Mainz investigate important scientific questions in Atmospheric Physics applying modern machine learning methods for big data sets.
In atmospheric physics, as in many other natural sciences, one is confronted with the situation that, due to the improvement of measurement technology and the enormous increase in computing power, huge amounts of data are available that can hardy be evaluated or not at all with conventional means. At the same time, we have a relatively poor understanding of the complex multi-scale system of the atmosphere; many fundamental processes and their impact on the system remain unclear.
Next to building up specialized machine learning methods, a major focus of the BINARY project is the development of new systems and infrastructure to handle the data size and throughput requirements.
Funded by Carl-Zeiss-Stiftung
Funding period: 03/2020 – 08/2025
Modeling across Multiple Scales and Disciplines
Computational methods and data-driven modeling have become indispensable tools across the sciences. The highly interdisciplinary Mainz Institute for Multiscale Modeling brings together researchers from different areas in natural and life sciences with researchers in mathematics and computer science. Our research follows two main thrusts: developing multiscale models informed by simulation and experiment, and pushing the boundaries of computational methods.
M3ODEL has been established in July 2019 as one of the Top-level Research Area funded through the Research Initiative of the State of Rhineland-Palatinate, and aims to facilitate and connect computational and modeling-oriented research across campus. It follows the path started by the center “Computational Science Mainz” (CSM).
Forschungsinitiative des Landes Rheinland-Pfalz
Investigating the topic of “Algorithmic Intelligence as an Emergent Phenomenon”.
Researchers from the institutes of Physics, Biology, Computer Science and the Max-Planck-Institute of Polymer Research at the Johannes Gutenberg-University Mainz study which statistical basic patterns are hidden in natural processes and how modern machine learning methods could acquire these patterns.
This research forms the basis for developing even better machine learning methods, and maybe even provides the opportunity to obtain a deeper understanding of the mysterious phenomenon called intelligence.
The High Performance Computing Research Group offers the following Lectures, Seminars, and Lab Courses:
In the Summer Term:
In the Winter Term:
