12th bwHPC Symposium

Thursday Sep 24, 2026, 9:00 AM → 5:00 PM Europe/Berlin

M629 (Universität Konstanz)

-- English version below --

Wir laden Sie herzlich zum 12. bwHPC-Symposium am 24. September 2026 ein. Die Präsenzveranstaltung wird von der Universität Konstanz organisiert.

Das bwHPC-Symposium bietet eine einzigartige Gelegenheit zum aktiven Dialog zwischen den Forschungsgruppen, den Betreibern der bwHPC-Dienste sowie den bwHPC-Support-Zentren. Im Mittelpunkt steht die Präsentation wissenschaftlicher Projekte und Erfolge, die mit Unterstützung der landesweiten bwHPC Hochleistungsrechner im Rahmen der BaWü-Datenföderation erzielt werden konnten. Die Teilnahme an dem Symposium ist kostenfrei und steht Wissenschaftlerinnen und Wissenschaftlern aller Fachrichtungen offen.

Alle Schritte zur Teilnahme am Symposium können auf dieser Webseite mittels des Menüs auf der linken Seitenleiste unternommen werden. Zur Einreichung von Beiträgen muss ein  (KIT-Indico-)Zugangskonto vorhanden sein oder erstellt werden, oder Sie schicken ihr Abstract (ca. 500 Wörter) mit Angabe des Beitragsformats an symposium2026@bwhpc.de. Die Verfasser*innen ausgewählter Abstracts werden dazu eingeladen kurze Publikationen als Beitrag für das Tagungsband zu erstellen, welches über das Konstanzer KOPS online veröffentlich wird. Alle Beiträge werden einen peer review Prozess durchlaufen.

Für weitergehende Fragen zur Veranstaltung stehen wir Ihnen gerne unter der Adresse symposium2026@bwhpc.de zur Verfügung.

Beitragsoptionen:

• Talk (~15 Minuten: ca. 12 Min Präsentation, 3 Min Diskussion)
• Poster

Wichtige Termine:

• Neue Deadline für die Einreichung von Abstracts: 24. August 2026
• Benachrichtigung über finalen Status der Einreichung bis spätestens: 04. September 2026
• Registrierungsdeadline: 14.September 2026

---

We cordially invite you to the 12th bwHPC Symposium on September 24, 2026. The event will take place on-site and will be organized by the University of Konstanz.

The bwHPC Symposium offers a unique opportunity to actively engage in a dialog between scientific users, operators of bwHPC services, and the bwHPC support centers. It's focus is on the presentation of scientific projects and success stories carried out with the help of bwHPC high-performance computing in the context of the BaWü data federation. The symposium will be free of charge and open to researchers from all scientific fields.

All steps for  participation at the symposium can be taken on this site using the menu on the left sidebar. For submitting a contribution to the symposium, you should possess or create an (KIT Indico) account, or you send your abstract (about 500 words) and type of contribution to symposium2026@bwhpc.de.Selected abstracts will be invited to write short papers for inclusion in the conference proceedings, published online via Konstanz KOPS.

 For further information on the event, please contact symposium2026@bwhpc.de.

Options for Contributions:

• Talk (about 15 minutes: approx. 12 min presentation, 3 min discussion)
• Poster

Key Dates to Remember:

• New Abstract Submission Deadline: August 24, 2026
• Notification of Acceptance Latest Until: September 04, 2026
• Registration Deadline: September 14, 2026

---

 

Abstract_Template(optional).txt

Arrival_Options.odt

Giesberg_Symposium.pdf

Grosser_Lageplan_05_2025.pdf

Hotels/Unterkünfte

Path_instructions_bwHPC_Symposium2026.odt

9:00 AM → 10:00 AM Registration

10:00 AM → 10:10 AM Address: Welcome Address

10:10 AM → 10:55 AM Talks

10:10 AM From light-field data to whole-brain functional ensembles: a GPU- and HPC-accelerated pipeline for larval zebrafish on bwForCluster NEMO2

Understanding how the vertebrate brain coordinates behavior requires recording its activity across the whole brain and at the speed of neural dynamics. We use light-field microscopy (LFM) to image behaving larval zebrafish volumetrically and at high frame rates, capturing calcium activity across the entire brain while the animal responds to visual stimuli and generates swimming, eye, and fin movements. Because every camera frame encodes a full high-resolution 3D volume, a single experiment yields hundreds of gigabytes of raw light-field frames.

We have built an end-to-end pipeline that runs on the bwForCluster NEMO2 in Freiburg:

(1) Volumetric reconstruction. Every raw light-field frame is deconvolved into a 3D volume by a GPU-accelerated wave-optics solver. We distribute this parallel workload across NEMO2's NVIDIA L40S GPU nodes (four GPUs per job).

(2) Preprocessing and registration. Reconstructed movies are motion-corrected and converted to ΔF/F. Each animal is imaged from a slightly different angle and position, and also has slightly different anatomical brain structures. Mapping brains into a common reference atlas allows us to perform volumetric statistical analyses across animals. These steps require large-memory CPU nodes (128 cores, 512 GB RAM).

(3) Decomposition and aggregation. Each recording is factorized by PCA-whitening plus independent component analysis (PICA) into ~2,500 spatiotemporal components — each a 3D spatial map with a time series. Components are then filtered, clustered (HDBSCAN and k-means) across the whole brain, and curated into a reproducible atlas of functional ensembles.

Speaker: Dr Sophie Aimon (Universität Konstanz)

10:25 AM CMF Fracture Segmentator: ein nnU-Net zur automatisierten Segmentierung von Frakturen des Unterkiefers

Einleitung
Die virtuelle OP-Planung (VSP) spielt eine zunehmende Bedeutung in der maxillofazialen Traumatologie. Voraussetzung dafür ist die Segmentierung von 3D-Bilddatensätzen zur Erstellung von 3D-Oberflächendaten. Es existiert bisher kein frei zugängliches System zur automatisierten Segmentierung und Fragmentseparierung von Unterkieferfrakturen anhand von CT-Datensätzen. Ziel ist es, durch den Einsatz moderner Machine-Learning-Algorithmen eine präzise, reproduzierbare und automatisierte Segmentierung von Frakturfragmenten bei Unterkieferfrakturen aus heterogenen bildgebenden Datensätzen (CT, DVT) zu ermöglichen.

Material und Methoden
Anhand bestehender, im Rahmen der klinischen Routine erstellter CT-Datensätze von Unterkieferfrakturen erfolgte unter Anwendung eines bestehenden neuronalen Netzwerkes zunächst die automatisierte Segmentierung des Unterkiefers. Anschließend erfolgte die manuelle Segmentierung der Frakturspalten, Aufteilung der Segmentierung in die Frakturfragmente und Klassifizierung. Die Segmentierung der Frakturen wurde für eine von zwei Untersuchern unabhängig durchgeführt und anhand von Dice-Koeffizienten und Oberflächendistanzen validiert. Anschließend wurde ein auf nnU-Net basierendes neuronales Netzwerk in der Erkennung, Kennzeichnung und Segmentierung von Unterkieferfrakturen anhand einer Teilmenge der Datensätze trainiert (Trainingsset). Im Anschluss wurde die Qualität der Vorhersagekraft des neuronalen Netzwerks anhand einer zweiten Teilmenge der Datensätze validiert (Kontrollset). Die Validität der Segmentierung wurde analog zur Analyse der manuellen Segmentierung untersucht.

Ergebnisse
Das Training des neuronalen Netzwerkes unter Verwendung der Rechenkapazität des HPC-Clusters des Landes Baden-Württemberg (bwUniCluster) anhand von 123 Trainingsdatenstäzen ergab für das Knochen-Label einen DSC von 0,95 (5-fold-Cross-Validation) bzw. 0,98 (keine Cross-Validation). Die automatisierte Segmentierung zeigte eine im Vergleich zur manuellen Segmentierung vergleichbare Ergebnisqualität (DSC 0,980 ± 0,004, HDavg 0,086 ± 0,156 mm, HDmax 2,079 ± 0,966 mm) in der Segmentierung von Fragmenten des Unterkiefergelenkfortsatzes bei gleichzeitiger Reduzierung des Zeitaufwandes auf wenige Minuten.

Schlussfolgerung
Die automatisierte Segmentierung von Unterkieferfrakturen mithilfe neuronaler Netze ist auf handelsüblicher Hardware und mit frei verfügbarer Open-Source-Software zuverlässig und niedrigschwellig umsetzbar. Eine klinische Anwendung dieses Verfahrens hat das Potential, die virtuelle OP-Planung bei der Versorgung von Unterkieferfrakturen entscheidend zu beschleunigen. Der Einschluss eines größeren, multizentrischen und multiethnischen Trainingsdatensatzes hat das Potential, die klinische Anwendbarkeit weiter zu verbessern.

Speaker: Johannes Schulze (Bundeswehrkrankenhaus Ulm, Deutschland, Klinik für Mund-, Kiefer- und plastische Gesichtschirurgie; Universitätsklinikum Ulm, Deutschland, Klinik für Mund-, Kiefer- und Gesichtschirurgie)

10:40 AM Global metagenomic reconstruction of the chicken gut virome reveals 19,778 viral populations and spatial partitioning

The gastrointestinal tract of chickens harbors complex microbial communities that play vital roles in host nutrition, physiological development, and immune regulation [1]. While bacterial microbiomes in poultry have been extensively characterized, the viral communities, or virome, remain poorly understood. However, the virome has significant potential to alter bacterial communities through predatory behavior or horizontal gene transfer. For example, phages can influence the abundance of key metabolic bacteria, thereby affecting the host's bacterial functional profile [2]. Therefore, studying virome dynamics is vital for a better understanding of the factors that contribute to poultry health and productivity and for potentially minimizing the amount of traditional antimicrobial additives used in poultry.
In this study, we conducted a large-scale global metagenomic exploration of the chicken gut virome. We combined our internal datasets with those available in public repositories and compiled a large dataset comprising 1,514 metagenomic samples (1,458 chicken gut metagenomes and 56 viral-enriched samples), originating from 15 countries. This dataset accounted for approximately 6 TB of raw data, with more than 50 TB of intermediate files created during processing. Bioinformatic analyses were initially performed on the BinAC cluster (University of Tübingen), followed by migration to BinAC2, which was crucial due to its larger storage capacities and faster performance. The analyses were performed as follows: raw samples were quality-controlled and filtered to exclude host DNA, then assembled into metagenomic contigs, which were used to recover viral operational taxonomic units (vOTUs). After dereplication and taxonomic annotation, vOTUs were used to build a comprehensive species-level viral catalog. We also performed additional downstream analyses to investigate virus-host interactions and to evaluate spatial diversity of phages across distinct geographical locations.
As a result, we first assembled 47,092 medium- to high-quality viral genomes and then successfully reconstructed a massive, non-redundant catalog containing 19,778 distinct species-level vOTUs. Taxonomic profiling revealed that most of them (97%) are double-stranded DNA (dsDNA) phages belonging to the Caudoviricetes class. Host-prediction analyses demonstrated that these phages primarily target bacteria from the gastrointestinal tract (GIT), including Lactobacillus, Limosilactobacillus, and Escherichia. We encountered difficulties during functional annotation, which was relatively inefficient, as most of the detected protein-coding genes remained uncharacterized and lacked known biological functions. The viral communities exhibited high specificity to the host, and, at the same time, a highly structured spatial stratification along the GIT. More precisely, the virome composition changed drastically between proximal and distal regions, with differences primarily driven by phages belonging to the Lactobacillaceae family.
Our findings demonstrated that the chicken gut virome is highly diverse, largely uncharacterized, and strongly localized along the GIT. It suggests that relying exclusively on fecal samples substantially misrepresents viral diversity. In addition, our findings contribute to our understanding of GIT microbial ecology and provide a basis for developing phage-based strategies to improve poultry health and productivity as an alternative to antibiotics.

Speaker: Timur Yergaliyev (Hohenheim Center for Livestock Microbiome Research (HoLMiR, 620))

10:55 AM → 11:30 AM Break + Poster Session: Coffee Break + Poster Session

10:55 AM Architecture and operational concept of a hierarchical storage management system across two sites for the provisioning of research data

The exponential growth of research data demands scalable and resilient
storage infrastructures that can span institutional boundaries while
maintaining seamless data accessibility. This work presents the design
and operational concept of a hierarchical storage management (HSM)
system deployed across two geographically distinct sites to support
collaborative scientific projects. This HSM solution demonstrates how
hierarchical storage can deliver high‑throughput, low‑cost data access
for multi‑site research collaborations.

Speakers: Iven Fellhauer (bwHPC - Heidelberg University), Sabine Richling

10:55 AM bwRSE4HPC – Research Software Engineering Support for High-Performance Computing in Baden-Württemberg

Research software has become an essential part of modern science, but many research groups struggle to develop and maintain code that is efficient, scalable, and sustainable. This is especially important as researchers turn to high-performance computing (HPC) systems to analyse increasingly larger datasets and perform more complex simulations. Common reasons for these include time constraints, a lack of software engineering experience, and frequent changes in personnel due to fixed-term research contracts. This can impact software performance and reproducibility, and can be detrimental to long-term maintainability of the software.

The joint initiative bwRSE4HPC was launched in 2025 to help address these issues. Being funded by the Ministry of Science, Research and Arts of Baden-Württemberg, the initiative provides researchers across the state direct access to a team of Research Software Engineers (RSEs) from the Scientific Computing Center (SCC) at KIT and the Scientific Software Center (SSC) at Heidelberg University. We work together with research groups to improve the performance, scalability, and usability of their software, with a particular focus on making effective use of HPC infrastructure. The initiative supports both experienced HPC users and researchers who are only starting to work with larger computing systems.

We provide support for short- and medium-term projects, usually lasting between 4 and 24 weeks. Depending on the needs of a project, this may include parallelization, GPU acceleration, profiling and performance optimization, improvements to build systems, language bindings, or automated testing. The support is free of charge and is adapted to the software, goals, and existing experience of each research group.

This poster gives an overview of the goals, organization, and scope of bwRSE4HPC and presents several software projects supported over the past year. The examples show the range of problems encountered in practice and the kinds of improvements that can be achieved through direct collaboration with RSEs.

Speaker: Glen Houston Hunter (SCC/KIT)

10:55 AM Direct Estimation of Information Shares in Higher-Order Continuous-Time Models

We propose a continuous-time framework to estimate time-invariant information shares. For this purpose, we employ a cointegrated CAR(k) process which accounts for the degree of temporal aggregation in the observable time series and avoids the many lags needed to specify discrete-time models for high-frequency data. We estimate the parameters by maximizing the Gaussian likelihood and investigate the statistical properties of the continuous-time information shares under mixed in-fill and long span asymptotics. We use simulations and an empirical application to show the benefits of continuous-time modeling for the analysis of price discovery in fragmented markets.

Speaker: Karsten Schweikert

10:55 AM Influence of phosphorylation on the structural stability of OTU11 studied with multiscale MD simulation

In plant cells, proteins marked with ubiquitin are degraded via the ESCRT machinery. This process is tightly regulated, for instance by ubiquitylating and deubiquitylating enzymes. One of these is OTU11, a deubiquitylating enzyme found in Arabidopsis thaliana [1]. Proteomics studies have identified six phosphorylation sites on the N-terminal frame of OTU11. Posttranslational modifications like phosphorylation can have variant effects on the function of proteins. To better understand these effects, the influence on the conformational ensemble of the respective proteins needs to be understood. We are interested in the impact of the phosphorylation on the structural stability of OTU11. Therefore we study the protein’s dynamics with a combined all-atom and coarse-grained approach.

[1] K. Vogel, et al. Lipid-mediated activation of plasma membrane-localized deubiquitylating enzymes modulate endosomal trafficking. Nat.
Commun. 2022, 13, 1, 6897. https://doi.org/10.1038/s41467-022-34637-3

Speaker: Madlen Malcharek (University of Konstanz)

10:55 AM Phenomic Selection in Sugar Beet: Optimizing Genotype Evaluation through Seed NIRS and UAV-Derived Field Emergence

Genomic selection has become a powerful tool in sugar beet breeding, but genotyping costs remain a barrier to routine, large-scale application. Phenomic selection, using seed near-infrared spectroscopy (NIRS) and UAV-derived field phenotyping as a cheaper alternative to marker-based prediction, is gaining attention, but rigorously benchmarking such pipelines across many genotypes, environments, preprocessing variants, and prediction models requires large-scale, repeated cross-validation, a computational workload well beyond desktop capacity.
In this study, conducted at the University of Hohenheim in collaboration with Strube D&S GmbH, we benchmarked seed-NIRS-based phenomic prediction for three sugar beet traits (corrected sugar yield, corrected root yield, sugar content) using data from 32 field trial locations across Germany, France, Belgium, and the Netherlands (2025 hybrid breeding program). We compared 190 NIRS profiles (raw + 189 Savitzky-Golay preprocessed profiles) across five prediction models (Elastic Net, LASSO, Partial Least Squares Regression, Ridge Regression BLUP, Random Forest), assessed the contribution of three UAV-derived field-emergence covariates (plant stand count, canopy cover percentage, plant instance area) to phenomic prediction accuracy, and modeled genotype-by-environment interaction under four cross-validation schemes (CV1, CV2, CV00, leave-one-location-out).
All computations ran on bwUniCluster 3.0 at KIT's Scientific Computing Center: more than 370 nodes, including 340+ CPU nodes (Intel Ice Lake, 64 cores/node; AMD EPYC 9454, 96 cores/node) and 28 NVIDIA A100/H100 GPU nodes, on a Lustre parallel file system. Screening 190 profiles × 5 models × 3 traits under 100 repetitions of 5-fold cross-validation produced roughly 1.4 million individual train-test model fits for the core phenomic prediction objective; evaluating the UAV covariates across 3 profiles × 3 models × 3 traits × 4 covariate configurations under 100 repetitions of 10-fold cross-validation added a further approximately 108,000 fits. Genotype-by-environment models were additionally fitted in a Bayesian RKHS framework (BGLR, 12,000 MCMC iterations, 2,000 burn-in per model) across three traits, two NIRS profiles, and four cross-validation schemes. We parallelized this workload as SLURM array jobs, submitting each profile-model-trait combination as an independent task, which made this scale of benchmarking practically feasible within the thesis timeline.
Savitzky-Golay preprocessing consistently improved predictive ability over raw spectra, with narrow window sizes and higher-order derivatives most effective; sugar content remained the hardest trait to predict. Among UAV covariates, plant instance area and canopy cover percentage improved predictive ability for yield traits, while plant stand count generally reduced it. Under the genotype-by-environment model, predictive ability stayed high when genotype or environmental information was available in training but declined substantially when both were absent.
These results show that seed-NIRS-based phenomic selection, supported by UAV-derived covariates, is a promising, cost-effective approach for genotype evaluation in sugar beet breeding. More broadly, they illustrate how bwHPC infrastructure was not merely a convenience but a critical enabler of this thesis: without large-scale parallel computing, systematically benchmarking hundreds of thousands of model fits across preprocessing profiles, models, and cross-validation schemes would not have been feasible.

Speaker: Jagadeeshwar Reddy Etukala (University of Hohenheim)

10:55 AM Protein stable isotope probing reveals anaerobic myo-inositol metabolism in the poultry intestinal microbiome

Myo-inositol (MI) is a biologically important cyclitol involved in diverse physiological processes in animals, yet its role as a substrate for intestinal microbial communities remains poorly understood. Although genomic studies have suggested that gut microorganisms possess pathways associated with MI degradation, experimental evidence demonstrating active MI utilization within the poultry intestinal microbiome is still limited. Understanding microbial substrate assimilation requires approaches capable of linking metabolic activity to specific microbial proteins and taxa. In this study, protein stable isotope probing (Protein-SIP) combined with high-resolution metaproteomics was applied to investigate anaerobic MI metabolism in the intestinal microbiota of laying hens.
Ileal and caecal digesta samples were incubated under anaerobic conditions in a customized minimal medium containing uniformly 13C-labelled MI as the primary carbon source. Samples collected at multiple time points were subjected to protein extraction and LC-MS/MS analysis to monitor isotope incorporation into microbial proteins. Metabolomic analysis of labelled metabolites was performed using nuclear magnetic resonance (NMR) spectroscopy to complement protein-based measurements. Identification of 13C-labelled peptides and assignment of microbial protein origins were performed using the SIPROS 4 workflow.
The computational analysis of Protein-SIP datasets represents a considerable challenge due to the increased search complexity associated with isotope-labelled peptides and the large volume of mass spectrometry data generated. In this study, 387 LC-MS/MS raw files with a combined size of approximately 566 GB were processed using SIPROS 4 through JupyterLab on the BinAC1 high-performance computing system within the bwHPC infrastructure. The available computational resources enabled efficient processing of the complete dataset and facilitated reproducible analysis of large-scale metaproteomic measurements that would have been difficult to perform using conventional local computing environments.
Protein-SIP analysis demonstrated active incorporation of 13C from MI into microbial proteins involved in energy production and conversion, carbohydrate transport and metabolism, and translation-related processes. Distinct metabolic patterns were observed between intestinal compartments, with the caecal microbiota showing earlier and stronger MI utilization compared with the ileal community. Taxonomic assignment of labelled proteins identified Megamonas as a major contributor to MI metabolism in the caecum, whereas isotope incorporation in the ileum was distributed across multiple bacterial groups, indicating differences in ecological niches and metabolic specialization along the intestinal tract. For the first time, this research highlights Gallibacterium as a candidate MI metabolizer in ileum.
This study provides the first experimental characterization of anaerobic MI metabolism within the poultry intestinal microbiome using isotope-resolved metaproteomics. Beyond revealing previously uncharacterized microbial functions, this work demonstrates the importance of scalable computational resources for analysing complex multi-gigabyte proteomic datasets. The integration of advanced mass spectrometry, stable isotope probing, and HPC-enabled data processing provides a framework for investigating microbial metabolism at community scale and highlights the role of shared research infrastructures such as bwHPC in enabling data-intensive life science research.

Speaker: Harshita Naithani (University of Hohenheim)

10:55 AM Support as a Service: Building the bwFDM-Helpdesk for and with the RDM community in Baden-Württemberg

The state initiative for research data management (RDM) in Baden-Württemberg bwFDM is advancing RDM support on the state level by implementing a new helpdesk system that will provide faster, more efficient support for researchers and staff at research institutions in Baden-Württemberg. Furthermore, the RDM community in Baden-Württemberg shall be actively involved in this process.

The bwFDM-Helpdesk aims primarily at academics and support staff working at research institutions that have no or insufficient in-house RDM support. Beyond that, the helpdesk is available to RDM support teams at universities who need assistance with questions beyond their own expertise or who seek state-level support for projects that align with the goals and objectives of Baden-Württemberg’s research data strategy and its implementation concept, both developed with bwFDM’s support.

The helpdesk’s previous system for handling enquiries was email-based and any information collected for the responses was stored via text documents in the bwFDM cloud. However, this approach was not scalable and the knowledge was not stored in a sustainable or easily searchable manner.

To mend these limitations, we implemented a ticket system and knowledge base. The active implementation into our workflow is currently an ongoing process. Both of these elements will enable us to cater to academics and support staff from research institutions in Baden-Württemberg in a more advanced form, allowing us to scale our support and making it more sustainable.

In order to pool the existing expertise within the RDM community in Baden-Württemberg, we are establishing a pool of experts. A survey is currently underway across the state, inviting people to join this expert pool. For example, anyone with expertise in specific RDM subject areas can sign up, so that when enquiries relating to these areas arise, we can contact the relevant experts from the pool and thus efficiently access expert knowledge.

Moreover, we are leveraging the joint support portal for research-related state services in Baden-Württemberg, hosted and technically supported at Karlsruhe Institute of Technology (KIT), making use of its centralized helpdesk platform currently also used by services like bwHPC, bwSync&Share and bwCloud. This portal has been migrated to the open-source software Zammad, which is also used by the EOSC-helpdesk and several NFDI consortia (e.g., NFDI4Chem, NFDI4Objects, GHGA). As they are based on the same software, it may be possible to connect the different helpdesks via technical interfaces, if this proves to be worthwhile in the future. By utilizing the joint support portal, bwFDM can offer general RDM-related consultations as a state service, which facilitates collaborations with other research-related state services. Along these lines, tickets can easily be processed in a joint fashion. Furthermore, the system offers an integrated knowledge base that can be built up and utilised by all participating services. Additionally, the portal’s connection to bwIDM means that future RDM support personnel at state level, such as data stewards at other research institutions or potential future stade-wide data stewards can be involved on a low-threshold basis.

Speaker: Béla Koch

11:30 AM → 12:10 PM Keynote-Talk

11:30 AM Neutron Star Dynamics and Gravitational Waves

We will present recent results in three areas related to neutron stars and gravitational waves, obtained through large-scale numerical simulations on modern high-performance computing systems:
A. The fully nonlinear evolution of neutron-star dynamics and the associated gravitational-wave spectra.
B. The dynamics of interacting neutron stars prior to merger, together with the emission of gravitational and electromagnetic waves.
C. The nonlinear dynamics of neutron stars in alternative theories of gravity.
These studies rely on large-scale three-dimensional numerical relativity simulations, advanced parallel algorithms to solve the coupled Einstein-hydrodynamics equations with high accuracy.

Speaker: Prof. Konstantinos Kokkotas (University of Tübingen)

12:10 PM → 12:25 PM Talks

12:10 PM Signoff Circuit Simulations for Modern Chip-Design using Leading-Edge Design Tools on the bwHPC NEMO2 Cluster

Due to high prototyping costs, modern semiconductor design relies on extensive signoff simulations to verify the electrical
performance and reliability of complex integrated circuits before production.
As technology nodes continue to shrink and circuit complexity grows, the computational requirements of these simulations can
become a significant bottleneck in the design flow.
Industry-standard design tools such as the Cadence Spectre Circuit Simulator keep up with this trend by introducing support for high-performance computing (HPC): Since 2019 parallelization on up to 256 CPU cores is supported, in 2023 the support of NVIDIA enterprise GPUs was introduced.

While commercial design companies typically can afford to own sufficient HPC resources to benefit from these features,
this is not feasible for small to medium-sized research groups in the field of microelectronics. The costs of purchasing high CPU and RAM compute nodes,
or enterprise grade GPUs as well as the effort in setting up and maintaining an automated workload distribution would not be justifiable for the one last
signoff simulation at the end of a design process. For those few occasions, the use of on-demand external cloud resources is way more reasonable as it allows to share HPC resources with other groups and access them only when needed.

A typical design flow in electronic design starts with short and interactive simulations on local workstations. The designers need immediate feedback to refine circuit parameters. This would not be suited for HPC workloads and queues. As a design evolves and is put together from many partial designs, grows bigger and simulations on multi-core workstations take too long. It is then desirable to use clusters like bwHPC NEMO2 and thus our group established a workflow for compute-intensive signoff simulations.

Scaling out to external cloud infrastructure with traditional load sharing paradigms requires a careful setup of the toolchain to meet challenges such as
different remote usernames, missing shared NFS file space, incompatible job scheduling, missing graphical remote desktop resources,
missing EDA software, missing PDK, license acquisition and non-disclosure agreements.
Additionally, we adapted to cloud specific constraints such as wall-clock-time limits or the need for regular two factor authentication.

By this we managed to successfully deliver our simulations to the bwHPC NEMO2 cluster and benefit from the large amount of high-end compute resources there. We observed simulation speed-ups of up to 2.1x / 2.5x on the CPU / H200 partition. GPU Cloud support was just recently released by Cadence and we expect to see even higher speed-ups there by fine-tuning the allocation of resources. The significant reduction of simulation time enables more comprehensive chip validation, allows to find design errors which we could not simulate before and improves the quality of our tape-outs.

Speaker: Johannes Stark

12:25 PM → 1:40 PM Break + Poster Session: Lunch Break + Poster Session

1:40 PM → 2:55 PM Talks

1:40 PM Scaling Extended Genetic and Evolutionary Algorithms on HPC-Systems: Master-Slave versus Island Algorithms

In this contribution we contrast parallelization of extended
genetic and evolutionary algorithms by global generational
parallelization (a master-slave architecture) with island
models (a message-passing architecture). The first class
of algorithms uses a single large population where mating of individuals
is unrestricted, whereas the second class of algorithms uses multiple
populations (called demes by biologists and used for the modelling of
ecological niches) where mating of individuals is restricted within a single
population and where a few individuals occasionally migrate between
neighboring populations.

As an introduction, we present a few innovations to both approaches:

• For the master-slave architecture, the degree of parallelization can be
  configured: The evaluation of the fitness function of a gene
  versus the complete genetic machinery except gene selection
  by pipeline compilation. In addition, various communication
  approaches between master and slaves can be configured.
  We concentrate on multicore and mpi based approaches.
• For island models, asynchronous parallel algorithms which
  communicate by message passing with the goal of using the
  assigned computational resources as efficiently as possible.
• Combinations of both architectures: Island models with islands parallelized by the master-slave approach.

For this purpose, the migration algorithms are highly configurable:
- The selection methods for emigrants as well as
for the genes which are replaced by immigrants.

• The communication topology (ring, 2-dimensional torus, 3-dimensional
  torus and generalized Petersen graphs, n random neighbors).

• Various communication strategies with and without synchronization.

• Self-adaptation of the number of generations.

• Support for configuring heterogeneous island models where
  each island runs an algorithm with a different configuration.

• The communication between processes is either by using a common file system (on a notebook) or by using mpi (on a HPC cluster).

Both approaches differ considerably with regard to communication:

The master-slave architecture copies all genes of a population once
per generation to the slaves. Its performance depends
crucially on the communication cost between master and slaves which
is usually acceptable for shared-memory multiprocessors.
The maximal number of genes is limited by the memory available
for the master process. However, exact replicability
of computational experiments can be achieved by proper configuration.

The message-passing architecture of island models has the advantage
that the maximal number genes scales with the number of cores (and
processes) available. The communication cost depends on the number
of genes which migrate between populations and is configurable.
Due to the inherent asynchronicity of the current implementation, exact
replicability of computational experiments must be replaced by
stochastic replicability.

We compare both approaches with regard to scalability and
to the speed of convergence of the resulting
extended and evolutionary algorithms.

We present a proof-of-concept implementation in the form of the
R-package xega, version 0.9.1.0 (https://CRAN.R-project.org/package=xega)
as well as a few results of computational experiments.

Speaker: Andreas Geyer-Schulz (Information Services and Electronic Markete (CIN-IEM), KIT)

1:55 PM Atomistic modelling of structures and processes in electrochemical energy storage and conversion

Processes in electrochemical energy storage and conversion are critical to mitigate the detrimental effects of global warming. We will show that atomistic simulations based on quantum chemical methods together with machine-learning approaches can contribute to improve, e.g., the ion mobility and stability of battery materials and the catalytic activity of electrochemical interfaces. Thus these simulations contribute to an accelerated materials design for improved electrochemical devices.

Speaker: Prof. Axel Groß (Universität Ulm)

2:10 PM Decoding the “sweet” talk: modeling the glycan code in bacterial infections

Bacterial infections pose a serious challenge due to antibiotic resistance; bacteria rapidly adapt to drugs used in clinical practice, resulting in high mortality rates. A deep understanding of the molecular mechanisms by which bacterial pathogens enter host cells provides a foundation for developing new, non-antibiotic strategies to combat these pathogens.
Cell surface is covered with complex carbohydrates (composed of several monosaccharides, i.e. glycans) that form an informational layer facilitating intercellular communication; in the course of acquiring pathogenicity, bacteria have evolved to recognize this host "glycocode". Experimental evidence indicates that lectins, carbohydrate-binding proteins, play a crucial role in the virulence factor arsenal of pathogenic bacteria, for example P. aeruginosa. They selectively bind to carbohydrate moieties of glycolipids and glycoproteins on the host cell surface. This binding induces complex changes in the state of membrane receptors within the membrane plane and triggers a cellular response that leads to the uptake or internalization of the resulting complex. Biochemical and biophysical methods are employed to elucidate the molecular mechanisms underlying the recognition of the host by pathogen virulence factors.
To decipher molecular aspects of the host recognition by virulence factors of a pathogen and bacterial uptake, e.g. identifying the membrane receptors targeted by a lectin and determining how lectin binding affects the composition of membrane domains, biochemical and biophysical methods, for instance fluorescent microscopy, mass spectrometry and immunoprecipitation, are employed. However, obtaining a detailed atomic-level information required for a profound understanding of complex formation within the context of the plasma membrane remains highly challenging. These difficulties stem primarily from the nature of glycans, which are characterized by structural heterogeneity and high intrinsic conformational flexibility.
Molecular dynamics methods offer a unique tool for tracking time-dependent changes in the conformational ensembles of aglycone-linked complex glycans, as well as their interactions with other plasma membrane components and bacterial lectins [1]. Furthermore, computational modeling enables the relatively rapid variation of conditions, allowing for the simulation of plasma membrane remodeling or alterations in glycosylation profiles.
In the current communication, we present examples illustrating the structural aspects of the interactions between P. aeruginosa lectins (LecA and LecB) with their glycolipids and glycoproteins receptors, thereby demonstrating the role of carbohydrate presentation on the plasma membrane surface in the processes of pathogen-host recognition [2,3].

References:
1. Perez S. & Makshakova O. Multifaceted Computational Modeling in Glycoscience. Chem. Rev. 2022, 122(20), P. 15914-15970. doi: 10.1021/acs.chemrev.2c00060.
2. Kociurzynski R., Makshakova O., Knecht V., Römer W. Multiscale Molecular Dynamics Studies Reveal Different Modes of Receptor Clustering by Gb3-Binding Lectins. J Chem Theory Comput. 2021, 17(4), P.2488-2501. doi: 10.1021/acs.jctc.0c01145.
3. Kittel A., Makshakova O., Hauerwas M., Edel N., Knickmeier N., Tomisch J., Aljohmani A., Yildiz D., Peyronnet R., Römer W. LecB from Pseudomonas Aeruginosa Modulates Piezo1 Currents and Localization in a Time-Dependent Manner. Cell Mol Life Sci. 2025, 82(1), P.399. doi: 10.1007/s00018-025-05934-z.

Speaker: Olga Makshakova (Synthetic Biology of Signalling Processes Lab, Signalling Research Centers BIOSS and CIBSS, Faculty of Biology, University of Freiburg, 79104 Freiburg, Germany)

2:25 PM Semiclassical Simulations of Dissipative Quantum Matter

Arrays of individually controlled atoms and ions have emerged as a leading experimental platform for both the exploration of complex quantum matter and the development of quantum technologies. However, guiding these efforts with reliable theoretical predictions is hard: the numerical cost of exact calculations grows exponentially with system size and quickly overwhelms even the largest computers. This challenge can be addressed by combining a semiclassical method, the truncated Wigner approximation, which takes leading-order quantum corrections into account, with large-scale parallel computation on CPU and GPU clusters. The approach replaces the full quantum evolution by a large ensemble of individual trajectories that can be calculated independently by solving coupled differential equations, making the problem naturally suited to high-performance computing.

The computational challenge arises from the combination of several requirements. The long-range interactions between the atoms couple every lattice site to every other, and the physics of interest unfolds over timescales orders of magnitude longer than the microscopic one, demanding many small timesteps per trajectory. Reliable estimates of observables in turn require averaging over large ensembles of trajectories run in parallel, placing significant pressure on memory. Carrying out these simulations on bwHPC resources allows us to access system sizes and timescales well beyond the reach of exact methods.

Using this setup, we study the dynamics of dissipative Rydberg atom arrays in one and two dimensions, where the interplay between dissipation and coherent evolution gives rise to rich nonequilibrium behaviour. We find that the strong interactions between atoms can dramatically slow down the approach to equilibrium, with relaxation proceeding through long-lived intermediate states and pronounced spatial anticorrelations that persist well after the initial configuration has been lost. These findings are directly relevant for current experimental efforts to realise and control strongly interacting quantum systems in the laboratory. The full results are presented in arXiv:2604.10538.

Speaker: Dr Viktoria Noel (University of Tübingen)

2:40 PM Multi-Scale Modeling of Electrochemical Interfaces and Processes

Batteries, fuel cells, photocells and many other applications are powered by fundamental electrochemical processes. Compared to surface science under UHV conditions, electrochemical systems combine a wide variety of additional effects. These range from the presence of an electrolyte and a multi-component environment to reaction conditions such as finite temperature, pressure, and electrode potential. Due to this complexity our knowledge of the ongoing processes is mostly limited to the macroscopic regime. However, nowadays interface-sensitive experiments together with theoretical modeling are able to provide deeper insights into structures and processes at the atomic level. This
fundamental knowledge makes the development and/or design of improved (electro)catalysts possible.
In this talk, we will compare the concepts of surface science and electrochemistry in detail, addressing both their similarities and differences. Using apparently simple electrocatalytic reactions as model systems, the effects of the reactive surrounding as well as environmental parameters will be successively explored. It turns out that pure and perfect catalyst models, which are often used in literature, are in many cases insufficient. Afterwards, these concepts will be extended from single crystals to the nanoregime, where nanostructured surfaces and particles are often used for electro-catalytic reactions. Taking transition metal alloys as an example, we will show that nanoparticles are not rigid objects but often change their morphologies and compositions under reaction conditions. Thus, understanding the dynamic nature of these catalysts is crucial in our efforts to further extend our ability to rationally design multi-component (electro-)catalysts.

Speaker: Timo Jacob (Universität Ulm)

2:55 PM → 3:30 PM Break + Poster Session: Coffee Break + Poster Session

3:30 PM → 4:30 PM Talks

3:30 PM Mining Multi-Omics Data for Hidden Mitochondrial Transcripts and Microproteins in Stress-Related Disorders

Mitochondria harbor a largely unexplored layer of genetic information beyond the 37 canonical genes currently represented in standard annotations. Recent studies have revealed that an increasing number of non-canonical mitochondrial transcripts and mitochondrial-derived microproteins (MDPs) can act as biologically active molecules involved in cellular stress responses, metabolic regulation, and resilience. However, these elements remain largely invisible to conventional transcriptomic and proteomic analyses, limiting our understanding of their contribution to human health.
Here, we establish an annotation-extended multi-omics framework to systematically uncover hidden mitochondrial transcripts and MDPs within existing large-scale omics datasets. Rather than generating new data, we leverage publicly available and in-house resources by expanding standard genome and protein references with custom-curated sequences representing known and predicted non-canonical mitochondrial elements. This enables their detection within established computational workflows using nf-core pipelines.
We apply this approach to stress-related psychiatric disorders, including post-traumatic stress disorder (PTSD), major depressive disorder (MDD), and related conditions, where mitochondrial dysfunction has emerged as an important contributor to altered cellular adaptation and stress resilience. Preliminary analyses from two distict PTSD studies demonstrate that extending conventional annotations enables the identification of additional mitochondrial transcripts and candidate MDPs that would remain undetected using standard workflows.
A major remaining challenge is the validation and genomic localization of candidate MDPs, as conventional annotation approaches do not reliably resolve non-canonical open reading frames within the mitochondrial genome. To address this, we will integrate publicly available ribosome profiling (Ribo-seq) datasets to provide orthogonal evidence of active translation and improve the assignment of mitochondrial-derived coding regions.
Together, this project establishes a scalable computational framework for exploring the hidden mitochondrial transcriptome and microproteome across large-scale omics resources. By integrating transcriptomics, proteomics, and future Ribo-seq analyses, we aim to generate a more complete understanding of mitochondrial contributions to stress biology. The presented workflow provides a reusable strategy for uncovering non-canonical mitochondrial elements across diverse disease contexts without requiring additional data generation.

Speaker: Sarah Viola Emser (Ulm University)

3:45 PM A standardized workflow for reproducible pig microbiome profiling across laboratories

The lack of standardized DNA extraction protocols in pig microbiome research limits reproducibility and cross-study comparability. This challenge becomes even more important in large-scale studies that generate substantial sequencing data and require coordinated, high-throughput computational analysis. To address this, we conducted a large-scale, multi-center validation study across six research facilities in Germany, evaluating three DNA extraction kits using both 16S rRNA gene amplicon sequencing and shotgun metagenomics on identical samples. Fecal samples from 200 male German Landrace × Pietrain pigs were processed independently at each site following harmonized protocols. Amplicon (V1–V2, V3–V4) and PCR-free metagenomic libraries were sequenced on Illumina NovaSeq platforms. A total of 7236 amplicon and 2700 metagenomic samples were generated, pooled and analyzed on the BinAC2 Cluster for bioinformatic processing. Analyses were performed using DADA2 and QIIME2 for amplicons and Trim Galore with moshpit distributions v2026.1 for metagenomes. Shannon entropy and Bray-Curtis distances were calculated as α- and β-diversity metrics.
Shannon entropy values derived from shotgun metagenomic profiles exhibited limited variability across facilities and extraction kits, with median values ranging from 3.42 to 4.09. For 16S amplicon data, entropy values were highly consistent across locations (median: V1-V2 6.93–7.80; V3-V4: 6.36–7.07), with narrow interquartile ranges and similar medians. Kruskal-Wallis tests for location were significant in nearly all comparisons (p<0.01), but numerical differences were small and not accompanied by clear shifts in diversity, indicating sensitivity to minor variation rather than biologically relevant effects. PERMANOVA on Bray-Curtis distances confirmed significant effects of location, extraction kit, and their interaction across all sequencing methods (p≤0.001 in all cases). Effect sizes were, however, minimal: location explained 1.25-2.4% of variance, extraction kit 3.6-10.1%, and their interaction with extraction kits ranged from 1.2-1.4%, with residual variation accounting for 87-93%. PERMDISP tests were largely non-significant, indicating that observed location effects predominantly reflect true differences in community composition (centroid position) rather than differences in within-group dispersion; a small number of kit/platform combinations showed significant dispersion effects and are interpreted with corresponding caution.
ANCOM-BC2 identified a subset of differentially abundant genera between locations within each kit and platform, consistent with expected facility-associated variation; the number and identity of these taxa varied by kit and sequencing method. This taxon-level variability did not translate into meaningfully altered overall community structure, underscoring that community-level consistency and individual-taxon sensitivity are compatible, not contradictory, findings.
Overall, despite minor statistically detectable compositional differences, microbial community structure was highly consistent across facilities and kits, with results showing that the selected DNA extraction kits, combined with the standardized protocol utilized, effectively reduce technical variance in multi-center pig microbiome studies. We propose this workflow as a standardized operating procedure for future porcine microbiome research, providing a reproducible framework that supports robust cross-study comparability and large-scale meta-analyses.

Speaker: Samuel Onyilokwu Enokela (Hohenheim Center for Livestock Microbiome Research (HoLMiR), University of Hohenheim)

4:00 PM How many zones does an electricity market model need?

The geographic scope of a reduced-scope electricity market model, how many zones it represents, and which ones, is a basic design choice. It is also a costly one: data preparation and simulation runtime scale sharply with the number of zones included, and can rise by several orders of magnitude when a full continental model is simulated in place of a single-zone one. Yet how this design choice propagates into simulated outcomes has not been systematically examined. This study provides the first such test, specifically asking whether the resulting scope sensitivity is driven by the number of zones simulated or by which zones are included, using a PowerACE campaign spanning connected-zone subsets (N = 1–48), seven bidding strategies, and three weather years. In total, more than 8,000 simulations are conducted, ranging from short runs of a few minutes to long runs of several days. Zone count has a statistically significant effect on simulated prices but explains only 30–60 % of scenario-to-scenario variance at fixed N. Classifying each scenario by the target zone's real, physically connected neighbours (Tier 1 of a graph constructed from simulated interconnector flows) recovers most of this variance on its own; including neighbours-of-neighbours or more distant zones adds little further explanatory power. The test applied here is deliberately conservative: complexity-matched linear and quadratic specifications, evaluated by cross-validated out-of-sample R², and required to meet the campaign's own precision target. Under this test, connectivity outperforms zone count in 54 of 70 statistically reliable cells (77 %), by a median margin of 20.2 percentage points of explained variance among the cells it wins. This advantage is not uniform. It is unanimous for well-connected zones (Germany, 12 real neighbours: 0 of 5 reliable cells reversed), but considerably less reliable at the periphery of the network, where zones with up to five real neighbours show reversals in 32–38 % of cells. Most reversals are attributable to a linear specification's failure to capture genuine curvature in the Tier-1 relationship, as confirmed by cross-validated quadratic terms. The single exception, Portugal, reflects a genuine but structurally narrow economic effect. The variable relevant to interpreting a reduced-scope model is therefore a target zone's real network embeddedness, rather than the model's overall size. The reliability of this diagnostic itself depends on how embedded the zone already is. A second outcome variable, intraday price variance, exhibits the same pattern, with an even stronger connectivity advantage (54 of 62 reliable cells, 87 %), indicating that the result is not specific to the average price level.

Speaker: Thorsten Weiskopf (Karlsruhe Institute of Technology)

4:15 PM Machine-learned intrinsic coordinates for vibrational calculations

We present a framework for molecular vibrational coordinate optimisation based on generative machine-learning methods, in particular normalising flows.
The computational cost of vibrational configuration interaction (VCI) calculations is largely determined by the size of the correlation space. We show that this space can be substantially reduced by optimising curvilinear coordinates variationally within vibrational self-consistent field (VSCF) or VCI theory. Starting from conventional Z-matrix valence coordinates, a normalising-flow transformation is trained to minimise the ground state vibrational energy or the sum of energies of selected excited states.
The resulting coordinates provide a more compact representation of vibration correlation and accelerate the convergence of subsequent VCI and perturbation-theory calculations. For H2CO, CH3F, and trans-HCOOH, the optimised coordinates reduce the correlation spaces required for converged excited-state energies by factors of 3-7. The reduction reaches several tens for strongly coupled states in dense spectral regions.
The framework is implemented in Python/JAX, with its core routines parallelised for both CPUs and GPUs. Although coordinate optimisation does not remove the exponential scaling of VCI, it substantially extends the range of accurate state-specific calculations that are computationally feasible.

Speaker: Andrey Yachmenev (Stuttgart University)

4:30 PM → 4:50 PM Address: Closing Address + Award Ceremony