| Room \ Time | Halle U1 | Hörsaal U1.101 | Seminarraum U1.191 | Seminarraum 02.073 |
|---|---|---|---|---|
| 09:00–09:30 | Coffee Reception & Registration | |||
| 09:30–10:30 | Welcome & Keynote Keynote lecture by Dr. Antica Čulina Title: Open science: between ideals and research realities | |||
| 10:30–10:45 | Coffee Break | |||
| 10:45–12:00 | Lightning Talks: Challenges, Ideas & Insights | Workshop 1: From Raw Data to a Website for Humans | ||
| 12:00–13:30 | Lunch Break | |||
| 13:30–14:00 | Panel Discussion: Making Data Curation Count
| |||
| 14:00–15:00 | Session 1: Curating Metadata | Session 2: Enhancing Reusability | Workshop 2: From RDM Classroom to Master’s thesis | |
| 15:00–15:30 | Coffee break | |||
| 15:30–16:30 | Session 3: Data Curation Workflows | Session 4: Collaboration & Sharing | Workshop 3: FAIR for Machines and Humans | |
| 16:30–17:00 | Conference Wrap Up |
Keynote lecture by Dr. Antica Čulina, Ruđer Bošković Institute, Zagreb
Title: Open science: between ideals and research realities
More information to follow.
9:30–10:30 | Hörsaal U1.101
How can we drive a cultural shift across the university to elevate the value and impact of data curation? Three distinguished professors discuss how data curation contributes to research excellence and what it takes to establish shared standards, strengthen institutional practices, and ensure that high-quality data curation is recognized and rewarded in academic career paths. The audience is invited to join the discussion.
Panel discussion with:
13:30–14:00 | Hörsaal U1.101
Pia Patricia Prkacin
Departement of Clinical Research, University of Basel / Division of Clinical Epidemiology, University Hospital Basel (USB)
10:45–12:00 | Hörsaal U1.101
In rural Lesotho, limited and unreliable internet access meant that study nurses in our multicentre antimicrobial stewardship study had to collect data offline using REDCap on their tablets. This approach came with an unexpected data quality problem: duplicate participant records.
Study nurses collected data offline during the day and synchronized the records with the REDCap server at the end of each working day when an internet connection was available. When the connection was poor, it was sometimes unclear whether a record had been successfully synchronized. Study nurses would therefore try to submit the record again later. Because participant IDs were generated automatically, a resubmitted record could receive a new ID and appear on the server as a different participant. Identifying and resolving these duplicates required time-consuming manual work. The duplicate records also created confusion among the study nurses when participants were due for follow-up three days after enrolment.
Once we understood what was causing the duplicates, our fix was deliberately low-tech. Study nurses began manually assigning each participant a unique ID at enrolment rather than relying on automatically generated IDs. Although this slightly increased the workload for them, it provided a simple and robust solution. If internet problems occurred again and a record had to be resubmitted, the same participant ID was retained, and the existing record was overwritten. This eliminated the problem of duplicates and simplified subsequent data cleaning and follow-up.
This lightning talk presents a practical lesson from a recently completed study rather than a technical innovation. In our setting, the manual solution was more reliable than the automated one. Our experience shows how a small decision at the point of data collection can create big problems downstream, and how a simple change in workflow can prevent them.
Anthea Alberto
Research and Infrastructure Support (RISE), University of Basel
10:45–12:00 | Hörsaal U1.101
While clean code is a very broad subject, this lightning talk will introduce the audience to the general principles as well as some concrete examples of it that can make a researcher’s life much easier. Code that is readily understandable and well-outlined documentation will reduce the effort that is needed to get (back) into a project, and is thus an important tool for data curation. Drawing on my long experience with process-oriented programming, both in my own research in political science as well as in implementation work for other scientists, I want to provide helpful input on best practices and useful tools for clean code and workflow documentation such as Markdown and Quarto.
Karen Kapur
10:45–12:00 | Hörsaal U1.101
University Children's Hospital Basel (UKBB)
Linking pediatric primary-care and hospital data across six healthcare institutions in Switzerland (SwissPedHealth-PREPP, an SPHN-affiliated project)
Multi-institutional clinical data linkage is often described as a technical problem: identifying common variables, privacy-preserving solutions, mapping them to shared standards, and connecting records across institutions. In practice, our experience suggests that an equally important challenge is understanding the context in which the data were created and transformed.
In a project linking pediatric primary-care and hospital data from three hospitals and three primary-care providers, we repeatedly encountered situations where correctly interpreting a variable depended on information not captured in the dataset itself: the interfaces through which clinicians enter information, institution-specific workflows, transitions between clinical information systems, and transformations applied as data move through extraction, de-identification, and harmonization pipelines distributed across sites.
For example, an apparent shift in a variable's distribution may reflect a clinical phenomenon — or a change in the information system behind it. The absence of a value from a structured field does not necessarily mean it was never documented. And when processing steps run at different institutions, small independent choices can interact in ways no single site can foresee.
Using practical examples, this lightning talk explores questions we believe are shared across disciplines: What contextual information is needed to interpret a clinical variable correctly? How should changes in source systems and workflows be represented in provenance? How can validation and inexpensive redundancy — variables captured at multiple sites purely for cross-checking — be designed into a project from the beginning, rather than reconstructed afterwards?
Data curation through shared, ongoing documentation of the systems, workflows, and transformations that give those values meaning succeeds best as a collaboration between researchers and data providers from day one.
Isabelle Marthot-Santaniello
Department of Ancient Civilizations (DAW), University of Basel
10:45–12:00 | Hörsaal U1.101
Since 2018, research has been ongoing at the Department of Ancient Civilizations on computational paleography of Greek papyri, first in the scope of D-scribes and now EGRAPSA projects. Some data have been published in various forms; others not (yet). Data collected and generated in the two projects are original digital images of the papyri, metadata (content, provenance, date…), and various annotations (image areas written by the same writers, parts of texts, characters…) or crops. In a time when “publish or perish” is often the tacit rule, when projects are evaluated mostly on their results published in journals or conference proceedings, it is challenging to determine the correct amount of time that should be spent on “publishing data only”: should it always go with a research article using it or is it worth publishing “data papers” in precise cases? How much effort should be spent on data curation and description? How can we check that we have not forgotten anything important? This presentation will use three datasets published in 2019, 2023 and this summer as examples to illustrate these challenges and the solutions (open to amendment) that were adopted.
Xeni Deligianni1, Saadia Ismail2, Michèle Leeman2, Ekaterina Stepanova2, Julia Anna Bielicki1
1University Children’s Hospital Basel (UKBB)
2sciCORE, University of Basel
10:45–12:00 | Hörsaal U1.101
Use Case: The SwissPedHealth National DataStream [Mozun et al.] was built within the Swiss Personalized Network (SPHN) [Lawrence et al.]. Routine clinical data from pediatric hospitals were transformed to graphs based on a Resource Description Framework (RDF) model, the main motivation being building a FAIR dataset that enables reuse of clinical data for research and quality assessment.
While this framework is especially set-up to increase compliance with FAIR principles, in practice, real world data curation and management comes with important challenges. In addition, RDF is a machine readable format but are we really ready to use it as it was meant? Is it adequate for helping the human curation? And how can such big and complex datasets be managed? Can AI be the answer or is it too soon?
Data curation: Curation of the graph database after delivery from the hospitals in order to deliver datasets for research and analysis is a complex task. We will shortly present the top challenges of working with a FAIR RDF format and the pipelines developed for evaluating the general quality of the data, adherence with cohort rules, applying irreversible anonymization, selecting project specific datasets for transfer and improving the metadata.
References:
R. Mozun et al., “Paediatric Personalized Research Network Switzerland (SwissPedHealth): a joint paediatric national data stream,” BMJ Open, vol. 14, no. 12, p. e091884, Dec. 2024, doi: 10.1136/bmjopen-2024-091884
A. K. Lawrence, L. Selter, and U. Frey, “SPHN - The Swiss Personalized Health Network Initiative,” Stud Health Technol Inform, vol. 270, pp. 1156–1160, Jun. 2020, doi: 10.3233/SHTI200344.
Eric Decker
Research and Infrastructure Support (RISE), University of Basel
10:45–12:00 | Hörsaal U1.101
The GPOS project website (gpos.omeka.net) is still online, but its core is gone. The film annotations, created for historical research, lived in an open-source video annotation database at Heidelberg University. That database was shut down years after the project left Heidelberg, about 15 years ago. Nobody had been designated as responsible for the project after its end, and its parts were spread across two infrastructures at different locations. At the time, research data management for "exotic" data such as digital film annotations was hardly established in the historical disciplines. Nor did anyone expect that the data could stay accessible apart from the software it was created in.
With Claude Code, I recently rebuilt a working viewer for the film and its annotations, and even added automated enrichment, in very little time. This was only possible because three things happened to be available: a private backup of the video, a private backup of the annotations, and my own knowledge as a former project member. Without any one of them, the information would most likely have been lost.
Yet the new viewer is not online. Building it was the easy part; deploying and running it is still unresolved. The university has the technical means, but it is unclear who takes responsibility the next time the presentation layer stops working. AI has made presentation software cheap to replace. It has not replaced data quality, documentation, or clear responsibilities. Who checks whether a presentation still works? Who replaces it when it fails? Who pays for that work?
This leads to the question I want to leave with the audience: if anyone can generate a custom interface for well-documented, high-quality data, should we still be operating long-term presentation layers at all?
Olga Serbaeva
Department Open Science, University Library Basel, University of Basel
10:45–12:00 | Hörsaal U1.101
We would like to invite you to a presentation of the Data Curation Guide, an interactive resource developed by the Research Data Management Network to guide researchers and professionals through data curation procedures, discipline-specific requirements and the research data lifecycle. The tool runs in the browser as a single self-contained file, with no account or internet connection required. It takes the discipline, the project stage and the data types as input and produces personalized guidelines as a PDF or CSV file. It also makes it easy to navigate some 400 data curation links and provides data stewards and researchers with a rich, fully interlinked glossary.
Andy Cao, Brigitta Gahl, Jeanne Lemant, Pia Neuschwander, Islam Salikhanov, Ingo Kutschka, Hassina Baraki
Department of Clinical Research (DKF), University of Basel
14:00–15:00 | Hörsaal U1.101
The University Hospital Basel is contributing to the European Association for Cardio- Thoracic Surgery (EACTS) Adult Cardiac Database, enabling benchmarking of surgical outcomes against other hospitals and countries. Our submission of over 10,000 surgical procedures will extend the database back to 2008, a substantial addition, since most of the currently available data starts from 2019 onwards.
The project was confirmed by the ethics commission to fall outside the scope of the Human Research Act and therefore did not require ethics committee approval. With data quality checks already complete, the remaining curation work centers on preparing sensitive clinical data so that it can be shared securely with external experts in medical research and quality assurance:
The anonymisation workflow is implemented in R using the sdcMicro package, allowing disclosure risk and information loss to be assessed iteratively rather than treating anonymisation as a one-time removal of identifiers.
The project illustrates a key challenge in clinical research data curation: balancing privacy protection, standardisation and data utility when making long-term historical data available beyond the institution in which they were collected. Beyond enabling this specific submission, the project strengthens the long-term curation and reusability of University Hospital Basel's cardiac surgical data: it increases the clinical utility of existing records, opens access to the EACTS database for Basel researchers going forward, and establishes a reproducible approach that could support similar data-sharing projects.
Deepak Unni
Swiss Institute of Bioinformatics (SIB)
14:00–15:00 | Hörsaal U1.101
The Swiss Personalized Health Network (SPHN) is a national data infrastructure coordinated by the Swiss Academy of Medical Sciences (SAMS) and the SIB Swiss Institute of Bioinformatics (SIB). The goal of SPHN is the standardization and harmonization of routine clinical data and make it available for research. Over the course of SPHN, numerous projects have generated valuable data assets. However, the reuse of these assets depends on their discoverability. Health datasets contain sensitive patient information and cannot be openly shared, they remain distributed across institutions, and information about their content, structure and access conditions has often been sparse or scattered. Without an openly accessible description of what is available, researchers cannot judge whether a dataset suits their question before entering the lengthy process of data access request.
The SPHN Metadata Catalog addresses this challenge by providing a centralized, curated description of data assets without exposing the underlying data. It publishes rich metadata for datasets originating from SPHN and related projects. Built on the FAIR Data Point specification, the catalog is aligned with Data Catalog Vocabulary (DCAT), and HealthDCAT-AP, an application profile developed within the European Health Data Space (EHDS) for describing health-related datasets, and extended with a lightweight SPHN Metadata Catalog Schema to address community-specific requirements while maintaining interoperability. In addition to descriptive metadata, the catalog provides quantitative dataset-level statistics expressed using the Vocabulary of Interlinked Datasets (VoID), derived as aggregated counts from the underlying datasets, typically represented as RDF knowledge graphs. The catalog also provides interactive visualization, through the SPHN Schema Scope, that further enables researchers to explore dataset structures, relationships, and the abundance of individual data elements, providing an overview of dataset content before requesting access. Metadata are available in both human and machine-readable RDF, with a SPARQL endpoint supporting programmatic discovery and querying.
We present the SPHN Metadata Catalog and its curation workflow, from template based metadata submission by data stewards to standardized RDF representation. We discuss lessons learned from harmonizing metadata across heterogeneous projects and demonstrate how standards alignment enables interoperability beyond SPHN.
Arnold Bosse
Research and Infrastructure Support (RISE), University of Basel
14:00–15:00 | Hörsaal U1.101
Researchers increasingly expect to query cultural heritage collections in natural language. The Model Context Protocol (MCP) lets AI assistants such as Claude access a collection’s data directly through “connectors”. Drawing on technology demonstrators built at RISE (University of Basel) [1] for the Dutch East India Company (VOC) archives [2] and the Rijksmuseum [3], this talk examines the curatorial roles AI plays in structuring, enriching and querying these collections.
For the VOC archives, the commodities thesaurus, which left most terms undefined, became a fully defined glossary. Definitions came first from project staff, the Woordenboek der Nederlandsche Taal and the Getty AAT. Where these had none, an LLM wrote them from corpus evidence, producing over half of the c. 3,500 entries, while recording its source and a confidence level. Researchers can now move from modern terms to period spellings, but the glossary partly reflects the model’s own reading of the corpus. For the Rijksmuseum, an AI-written parser turned semi-structured provenance text into c. 100,000 structured events, queryable collection-wide by collector, price or transfer. Its event model follows established museum provenance standards (AAM, CMOA), but an LLM decided how ambiguous phrasing fits that model.
Beyond the data, an MCP server extends curation to the means of querying it. Which tools exist, what they filter on and how they are described all steer an AI agent, much as a finding aid steers a reader in the archive. Skill files go further: written in natural language, they can tell an agent how to read and evaluate a collection, giving an institution a direct say in how its holdings are interpreted.
If institutions do not write and share the tools and skills through which agents read their collections, others will, bringing their own assumptions and priorities. What would institutions gain in authority and impact by taking on this role deliberately, and what might they lose in openness to other readings and collection interfaces?
[1] https://rise.unibas.ch/
[2] https://github.com/kintopp/globalise-mcp/
[3] https://github.com/kintopp/rijksmuseum-mcp-plus/
Malin Ziehmer-Wenz, Sumin Kim, Thiery Masserey, Nakul Chitnis
Swiss Tropical and Public Health Institute (Swiss TPH)
14:00–15:00 | Seminarraum U1.191
Swiss TPH’s storage systems and archives extend far into the past: datasets collected decades ago on paper, digitized with software no longer in use, and stored in file formats that are not easily converted. Their documentation is a few codebooks, some folder names and the memory of colleagues who have since moved on. One example is the Kilombero Malaria Project, whose field data were collected in the 1990s and used for malaria modelling before being archived. Over the past two decades, OpenMalaria has continued to evolve, incorporating new features and research objectives. This creates new opportunities to revisit these historical datasets and use them to parameterize the model for questions they could not originally address.
This lightning talk outlines our procedure from archive to publication and how AI can support in locating and recovering the data, establishing documentation and metadata, converting to open and durable formats, checking applicable legislation, contracts and consent for what allows sharing, and finally publishing.
Along the way we discovered that the hardest questions are often not technical: data collected long before today’s rules frequently have no recorded consent scope, legal basis, contractual terms or ownership, and each answer decides whether a dataset can be shared at all.
Takeaway: raw data is not enough. Data becomes valuable once it is documented, stored in open and durable formats, and cleared for safe sharing. Despite all these challenges, unearthing this gold is well worth the effort: these datasets cannot be easily recreated, and making them reusable gives them a second life for future research.
Soheila Aghlmandi
Paediatric Research Center, University Children's Hospital Basel (UKBB)
14:00–15:00 | Seminarraum U1.191
Use case. Swiss health insurers record every reimbursed consultation, drug, laboratory test and hospital stay, but these claims were built for billing, not research: they contain no ambulatory diagnoses and no patient identifier shared across insurers or with clinical cohorts. We curated claims from up to three insurers (35–50% of the Swiss population) for three studies: a nationwide randomised trial of antibiotic prescribing feedback in 3,426 physicians (JAMA Intern Med 2023), the reuse of that trial cohort to measure the impact of COVID-19 on prescribing (Clin Microbiol Infect 2024), and a privacy-preserving linkage of claims with the Swiss HIV Cohort Study to analyse costs of care (Open Forum Infect Dis 2025).
Curation aspects. (1) Standardised extraction: insurers' data managers applied a common protocol and pseudonymised physician and patient identifiers at source, so the same pipeline could be rerun after the trial ended. That turned a closed trial dataset into a 48-month cohort. (2) Documenting denominators: a published commentary compared our “antibiotics per 100 consultations” with sales per inhabitant and with sentinel networks, and drew different conclusions. The data were not wrong; the definition of the population at risk had not travelled with them. We argue that numerator, denominator and unit of aggregation belong in the mandatory metadata. (3) Sensitive-data linkage: dates of birth were encrypted with Bloom filters by an independent statistician. Linkage scenarios were compared against the full cohort to document representativeness, at the cost of linking only a third of records. (4) Access: anonymised data can be shared only with approval of the funder's research committee and each insurer.
Dead ends and open questions. Claims arrive six to nine months late. Comorbidity must be inferred from drugs. Every agreement was negotiated project by project. We will discuss what a reusable claims-data package should contain (data dictionary, extraction code, denominator definitions, linkage provenance) and whether AI-assisted inference of diagnoses from billing codes can be curated transparently enough to be trusted.
Julien Roux
Department of Biomedicine, University of Basel
14:00–15:00 | Seminarraum U1.191
Single-cell and spatial transcriptomics technologies have transformed the study of biological systems and disease mechanisms. One challenging step in analyzing such high-dimensional data is accurately identifying cell types within a new dataset. Most approaches rely on comparing cells to collections of annotated references, which are unfortunately scattered across multiple repositories and are heterogeneous in terms of format and annotation quality. Finding, understanding, and reusing the suitable references for a given project can be as time-consuming as the analysis itself.
We have recently developed a structured database designed to centralize and manage diverse reference datasets, originating from multiple omics technologies. These datasets can come from public repositories or controlled-access sources. Our aim was to simplify the submission process as much as possible to encourage contributions from the community, while ensuring careful standardization and versioning of the datasets and their metadata to enable full reproducibility of the analyses.
Building on this foundation, we would like to focus now on developing a suite of advanced tools that:
(i) Harmonize free-text annotation of reference datasets using ontologies, controlled vocabularies, and LLMs.
(ii) Enhance data accessibility and usability, notably by providing intuitive search capabilities and visualization
(iii) Provide data-driven suggestions to the users for the most accurate reference datasets to use for their analysis.
This work should ultimately improve the discoverability and interoperability of reference datasets. For the genomics community, this addresses a key bottleneck in many projects involving single-cell data.
Valentin Schittny
Department of Pharmaceutical Sciences (DPhW), University of Basel
15:30–16:30 | Hörsaal U1.101
Experimental gene-therapy research generates heterogeneous data across different instruments, file formats and storage locations. Scientific insight often emerges only by combining measurements acquired at different stages of an experiment, making consistent identification, organization, metadata and traceability essential. During my PhD, I therefore developed a lightweight and modular data pipeline designed to reduce repetitive data handling and make everyday experimental work easier.
The pipeline is organized around a defined project structure and unique experimental identifiers. Starting from an experimental ID, it locates and retrieves associated raw data from multiple sources while preserving the original files, assigns them to a standardized project structure, executes the relevant analysis workflows and generates standardized outputs. Experimental metadata, results, figures and contextual information are subsequently assembled into an automated HTML report, creating an experiment-level record that can be understood and revisited by “future me” or other researchers. Together, these steps provide a consistent path from distributed instrument outputs to documented experimental results while reducing manual work and the associated risk of copy-paste, transcription and file-handling errors. Because analyses and reports can be rerun, corrections to the workflow can also be propagated consistently to previous experiments. Code versioning, reproducible software environments, synchronization and backup strategies further support traceability and portability.
The basic architecture was established during the first month of my PhD and has remained in active use for approximately three years. Rather than following a fixed software specification, the pipeline has evolved bottom-up as new instruments, file formats, analyses and programming languages were incorporated, while its core structure remained largely unchanged. Although it was not originally designed around formal FAIR or research-data-management principles, many related practices emerged naturally from solving recurring problems in everyday research.
This use case demonstrates how data curation can become an integral part of active experimental research rather than an additional task performed at its end. The aim is not automation for its own sake, but to reduce repetitive manual work while improving structure, traceability and reproducibility — ultimately making the correct workflow the easiest workflow.
Wang Zhao, Pei Shan Loo, Daniel Cobos
Swiss Tropical and Public Health Institute (Swiss TPH)
15:30–16:30 | Hörsaal U1.101
As public health modelers at the Swiss TPH, we fit one simulation model of chronic disease to data from more than 200 countries and cities. Like many researchers, we write code on a laptop, often with AI coding assistants, and run our analyses mainly on sciCORE, the shared university computing cluster.
Our everyday practice seemed harmless: before each run, we copied the latest scripts and input files to the cluster, letting jobs write directly into the result folder. Every computation relied on live, moving files rather than fixed snapshots.
The resulting problems were quiet but severe. Edited scripts failed to transfer, leaving older code to execute undetected; stale outputs from previous runs were mistakenly reported as fresh results; and worst of all, most of these failures raised no error: the jobs finished while silently producing wrong results.
Our hard-learned lesson was that spot-fixing mistakes cannot cure a vulnerable workflow. When code and data change continuously, isolation is needed for auditability and reproducibility, and every computational run must carry everything it needs.
The solution is what we call a "cabin": a sealed, self-contained run folder named by model version, setting, and date. A cabin freezes copies of the scripts, the simulation model and its input files, records the Git commit hashtag, and stores checksums of the files. Before computing, it checks itself against these fingerprints and halts immediately if anything has changed. Results are stored inside each cabin instead of a shared folder, so that they can be retrieved together with the exact files that generate them.
This workflow directly addresses core data curation challenges: computational reproducibility, provenance tracking, automated verification, and long-term archiving. It serves everyone involved in the research lifecycle: investigators get readable audit logs, AI coding assistants receive written rules on how to build and use cabins, reviewers can audit re-runnable analyses, and the self-contained cabins are ready for long-term preservation.
Mehrdad Davoudi1, 2, Jacqueline Romkes1, Michèle Widmer1, Chris Easthope Awai3, 4, and Elke Viehweger1, 5
1Department of Neuroorthopaedics, University Children’s Hospital Basel (UKBB)
2Clinic for Orthopaedics, Heidelberg University Hospital
3Data Analytics and Rehabilitation Technology DART, Lake Lucerne Institute
4Department of Health Sciences & Technology, ETH Zurich
5Department of Biomedical Engineering (DBE), University of Basel
15:30–16:30 | Hörsaal U1.101
This presentation addresses the use of generative large language models (LLMs) in clinical biomechanics, paediatric neuro-orthopaedics, and gait analysis, focusing on the curation of kinematic data for transparent and reproducible AI-assisted interpretation. The work is based on GaitSharing, an open-source data management and processing system for organising, analysing, and preparing gait data for clinical research.
Three-dimensional instrumented gait analysis supports surgical decision-making in children with cerebral palsy (CP), but its interpretation is time-consuming and subject to inter-rater variability. In this single-centre pilot study, we investigated whether an LLM could generate gait-deviation findings and surgical procedure suggestions that align with expert judgement.
As a clinical use case, lower-limb kinematic features from eight children with unilateral CP were extracted using GaitSharing and submitted three times per patient to OpenAI’s GPT-5.5 model through a structured prompt. The data-curation workflow included anonymisation, transformation of kinematic curves into structured features, pairing with reference values, prompt standardisation, and documentation of repeated model outputs. The model assessed 28 kinematic deviations and 12 surgical procedure groups using majority voting. One gait analyst and two paediatric orthopaedic surgeons independently evaluated the outputs.
Agreement with the gait expert was highest at the hip (90.6%) and lowest at the knee, particularly in the transverse plane (65.2%). Agreement for surgical procedures reached 83.9% and 73.4% for the two surgeons. Inter-surgeon agreement was 79.2%, with a weighted Cohen’s kappa of 0.59, indicating moderate agreement. The LLM showed high self-consistency (>90% across runs).
These preliminary findings demonstrate how an open-source data management system can support anonymisation, structured feature extraction, provenance documentation, reproducibility, and responsible reuse of sensitive clinical gait data. They also suggest that LLMs may serve as assistive tools, although confirmation in larger and more diverse cohorts is required.
Open-source toolkit: https://github.com/MehrdadDavoudiii/GaitSharingToolkit
Documentation: https://mehrdaddavoudiii.github.io/GaitSharingToolkit/
Publication: https://www.mdpi.com/2306-5354/13/8/862
Margareta Scherr1, Shauna L. Rohner², Nina Funke2, Myriam V. Thoma1
1Faculty of Psychology, University of Basel
2School of Health Sciences, Eastern Switzerland University of Applied Sciences (OST)
15:30–16:30 | Seminarraum U1.191
The study Ageing in the Aftermath of Adversity follows up individuals at high risk of early life stress (ELS) exposure under the former Swiss system of compulsory social measures and placements (CSMP), compared with age-matched controls. Extending an existing longitudinal cohort, this study aims to (1) compare long-term health outcomes in older adults with differing ELS experiences and (2) examine dynamic interactions among biological, psychological, and social factors and their predictive value for ageing trajectories. The extended follow-up allows assessment of late-onset outcomes, time-varying and cumulative effects, and potential critical periods along trajectories that lead to resilience. In this follow-up, biomarkers will be added to test whether traumatic stress is associated with chronic changes in HPA-axis activation, including salivary cortisol and alpha-amylase, and nail cortisol and cortisone.
In theory, integrating clinical psychology with biological measurements offers a more comprehensive approach to understanding ageing trajectories. In practice, however, this integration brings together two research cultures with different terminology, methodological routines, and implicit knowledge. Bridging these disciplines requires an understanding of relevant pre-analytical procedures, appropriate sample storage conditions, and the practical challenges of incorporating biomarkers into clinical research settings. Gaps in current and future research may arise not only from differences in methodological expertise but also from inadequate data documentation. Information routinely recorded in one discipline may be overlooked in another.
The aim of this presentation is to address learnings from planning an interdisciplinary study: (1) how to reconcile the different requirements of two disciplines in joint data collection and (2) how to improve communication with laboratories. Presenting a study in the middle of its planning stage, this contribution also aims to openly discuss the implementation and improvement of interdisciplinary teams and the curation of a shared language for data.
Ricardo Diogo Righetto
Biozentrum, University of Basel
15:30–16:30 | Seminarraum U1.191
Cryo-electron tomography ("cryo-ET") is an imaging technique for capturing the internal cellular environment at sub-nanometer resolution. Thanks to advances in instrumentation and sample preparation, as well as in the computational hardware and software, the method has gained wide popularity among structural biologists in recent years. Cryo-ET presents the unique advantage to directly observe protein complexes in their native context, providing insight into how they interact with other and to other cellular entities such as membranes, biomolecular condensates and organelles. Each dataset is a treasure trove of biology containing information often exceeding the scope and analysis capability of single research groups. However, the large volumes of data being generated and the lack of standards for deposition and information exchange in the field make it very hard for scientists to explore and reuse cryo-ET data in practice. Even tools and guidelines for the curation and systematic cataloging of our own data are scarce. Furthermore, deposition of raw and intermediate data on current public repositories is cumbersome and not enforced by journals, such that scientists feel they have little incentive to actually do it. In this talk, I will present the experience of our group in sharing cryo-ET data, with emphasis on a collaborative effort between academia and industry to generate a large dataset of Chlamydomonas tomograms as a community resource. I will present the challenges and rewards in doing so, and what I consider is still needed to make cryo-ET data more democratic and efficient through data sharing.
Georg S. Bailer, Christoph R. Meier, Cornelia Schneider
Department of Pharmaceutical Sciences (DPhW), University of Basel
15:30–16:30 | Seminarraum U1.191
Sex is one of the most frequently collected variables in research with human data, but how do you use, interpret and report it? Health sciences have adopted new reporting guidelines (e.g. SAGER) and funders as well as journals increasingly expect sex- and gender-disaggregated analyses. As a result, this ubiquitous variable now appears in a growing number of analyses, but do the underlying data actually support the analyses built on them?
Drug safety research relies heavily on the secondary use of data, which is only meaningful if the data are well documented and curated. At the Basel Pharmacoepidemiology Unit we work with Swiss claims data, primary-care records, a hospital data warehouse, cohort data and survey data. Across all of them, “sex” looks like a simple and interchangeable variable, and precisely because it looks simple, it is rarely curated. The information is entered by an insurer for billing, by a clinician, or by a participant on a questionnaire, and then travels unexplained and unchanged into every derived dataset without further documentation and reassessment. The label is the same everywhere; the concepts behind it are not. The uniform interpretation conceals heterogeneity and by the time a paper reports that “women had a higher risk”, nobody can say with certainty who these women were.
Drawing on our drug safety studies, we present pitfalls when trying to understand and interpret “sex-based results”, structured along three questions:
Lars Schöbitz
Global Health Engineering, ETH Zurich
10:45–12:00 | Seminarraum 02.073
Abstract
Survey and monitoring data from the water, sanitation and health sector mostly lives in spreadsheets that only their owners can read. For five years, Global Health Engineering at ETH Zurich has turned such files into published data. Each dataset becomes an R data package with a data dictionary and metadata, archived with a DOI and presented on a website with documentation and plots. AI now does most of the mechanical work. We publish for machines and humans. This workshop is about the humans, who still have to discover, interpret and trust the data.
In this workshop we build one dataset together. Every participant contributes to it through GitHub, a little before the conference and the rest in the room, and the contributions arrive as different kinds of raw data. I then run the workflow live with the free, open source tool washr, from these raw files to a data package and its website. A coding agent performs the steps. The points where a person confirms stay visible, and so does the record of what the AI did and of who contributed what. Participants review the documentation and propose changes on GitHub. In groups we then discuss four themes the workflow raises for data curation: oversight of AI curation, disclosure of AI contributions, sensitive data, and scientific credit for data.
Participants who could imagine publishing a dataset of their own this way can tell me after the workshop, and I follow up with them.
Agenda
Learning objectives
Participants will be able to:
Prerequisites
Sarah N. Musy, Tania T. Martins, Franziska Zúñiga, Michael Simon
Nursing Science, Department of Public Health, University of Basel
14:00–15:00 | Seminarraum 02.073
Abstract
Good data curation practices should become part of how researchers work with data from the outset. But how can we introduce these practices during graduate education without overwhelming students with requirements designed for larger research projects?
At the Institute of Nursing Science, our approach to data-related education has evolved from teaching reproducible statistical analysis with R to introducing dedicated research data management (RDM) teaching in 2023. Our current challenge is moving from teaching RDM principles to supporting their application in students’ own projects. We are therefore developing practical resources for Master’s students and their supervisors, starting with a short Data Management Plan and simplified informed-consent guidance and templates. Further resources may follow as needs are identified.
In this interactive and interdisciplinary workshop, we will present our approach and initial resources as a starting point for discussion. Working in small groups, participants will identify challenges faced by Master's students, explore what practical support they need, and propose resources that could help them manage their data responsibly. Each group will prioritize its ideas and outline the purpose and content of one proposed resource.
Together, we aim to develop an initial collection of ideas for a practical student RDM toolkit, while exploring which needs are shared across disciplines and which require discipline-specific approaches.
Using nursing and health sciences as a use case, the workshop asks a broader data-curation question: How can we turn RDM education into everyday research practice from the beginning of a researcher’s career?
Agenda
Learning Objectives
By the end of the workshop, participants will be able to:
Prerequisites
No prior RDM expertise or preparation is required. Participants from all disciplines and at all career stages (students, PhD students, researchers, lecturers, professors, etc.) are welcome. Bringing a laptop or smartphone is recommended for contributing to the collaborative exercise.
Noémi Villars, Daniela Subotic
Swiss National Data and Service Center for the Humanities (DaSCH)
15:30–16:30 | Seminarraum 02.073
Abstract
Funders and journals now expect research data to be FAIR, and most of what that requires comes down to information only researchers can provide. What happens when that information is missing, and what do researchers gain when it is there? This workshop explores both sides in practice.
In research areas where data is often an interpretation of physical sources, reusers need more than a title and a licence. They need to know which sources the data draws on, how these were transcribed or classified, and what is uncertain. Without proper documentation, reusing a dataset can be difficult, if not impossible.
The session is hands-on. Participants work with two versions of our showcase project "Alice in DaSCHland": the full project and a deliberately stripped-down copy with no licence, vague labels and no description of the sources. We try to answer everyday questions on both versions: What does this data mean? Where does it come from? May I use this image in my own publication? Where the answers stop, we trace which FAIR principle is affected and what that means for the researchers behind the data: not being found, not being cited, or being misread.
In a second step, participants who have brought their own data (recommended but not mandatory) can check it against the same three questions. Together we turn the findings into a short checklist of what is to be done with their data to make it FAIR(er) that the participants can take home.
We don't have all the answers ourselves and look forward to the discussion.
Agenda
Learning Objectives
Participants will be able to:
Prerequisites
Quick Links