FAIRmat participated in and co-organized a Research Data Management (RDM) Workshop for the DFG Research Unit "Energy Landscapes and Structure in Ion Conducting Solids (ELSICS)". The workshop took place at Marburg University from October 9 to 10, 2025, in collaboration with NFDI4Chem and ELSICS.

The program began with an introduction to the fundamentals of RDM, followed by presentations from NFDI4Chem and FAIRmat. Our team presented the NOMAD ecosystem, highlighting how it supports the management, sharing, and reuse of research data in condensed-matter physics and materials science. Afterwards, ELSICS spokesperson Karl-Michael Weitzel provided an overview of the research unit and its ongoing projects. Researchers from various ELSICS subprojects shared insights into their work and the data they generate, including file formats, data volumes, and the tools they use. Their contributions offered a valuable overview of the consortium’s data landscape and helped shape tailored recommendations for best practices.

The second day featured four thematic workshops on key aspects of RDM: selecting a repository, the advantages of using electronic laboratory notebooks (ELNs), standardization and metadata, and collaboration and data sharing. The discussions focused on how these topics can improve scientific workflows, enhance reproducibility, and promote the adoption of effective, sustainable data management strategies.

The workshops showcased how NOMAD Oasis can facilitate collaboration within research consortia by enabling shared access to data, metadata, and experimental samples under well-defined conditions. For the FAIRmat team, the event provided an excellent opportunity to better understand the RDM needs and practices of research units like ELSICS, thereby strengthening the dialogue between infrastructure providers and the scientific community.

We are proud to announce that FAIRmat has reached a major milestone in data standardization. In close collaboration with the NeXus International Advisory Committee (NIAC), our experts have extended the NeXus data format – originally developed for X-ray, neutron, and muon science – into a cross-domain standard for experimental data in materials science.

The latest update introduces new application definitions for:

• Atom Probe Microscopy (NXapm): standardized data exchange for atom probe experiments.
• Electron Microscopy (NXem): a unified standard for scanning and transmission EM and focused-ion beam capabilities of such instruments.
• Optical Spectroscopy (NXoptical_spectroscopy, NXellipsometry, NXraman): a common framework for experiments with tailored optical setups.
• Photoemission Spectroscopy (NXmpes, NXmpes_arpes, NXxps): multidimensional data support across different photoemission instruments.

Additional base classes also improve the description of experimental geometries and component calibrations.

NeXus development follows a transparent, community-driven process: new definitions are openly discussed and refined on GitHub before being adopted as official standards. FAIRmat's sustained contributions have been essential in shaping this process, ensuring that NeXus remains both domain-specific and interoperable across disciplines.

These new definitions are already fully integrated into our research data management platform NOMAD, allowing researchers to upload, explore, search, and share NeXus-formatted data directly on the platform. NOMAD preserves the structural semantics of NeXus and makes them interoperable with other materials data through its schema-based Metainfo system.

By combining standardization efforts in NeXus with the powerful infrastructure of NOMAD, FAIRmat helps researchers with agreeing on how data should look like and working with these in practice.

We are excited to share that a new FAIRmat paper has been published in Nature Communications! In the article, titled "Towards an interoperable perovskite description or how to keep track of 300 perovskite ions", the authors propose guidelines and a standardized JSON schema for reporting perovskite compositions.

Hybrid perovskites are a cornerstone of modern optoelectronics, but their vast compositional flexibility has led to confusion and inconsistencies in the way materials are described. With the increasing importance of machine learning and data mining tools, ensuring computational readability has become an important concern.

To address these challenges, the work introduces a JSON schema that follows IUPAC recommendations and enables both human- and machine-readable descriptions of perovskite materials. This schema captures descriptors including composition, molecular formula, SMILES representation, IUPAC name, and CAS number for more than 300 identified perovskite ions.

In addition, the team provides utilities that automatically generate standardized descriptions from common abbreviations and coefficients, as well as a curated database of all identified ions. The Hybrid Perovskite Ions Database is available via the NOMAD API, enabling researchers worldwide to upload, share, and reuse consistent perovskite data in line with FAIR principles.

The proposed approach not only eliminates ambiguities in reported perovskite compositions but also enables better use of the rapidly growing body of perovskite data for machine learning and computational studies.

On September 11-12, FAIRmat hosted a successful on-site part of the 3rd LLM (Large Language Models) Hackathon for Applications in Materials Science and Chemistry, welcoming participants from across Germany. This was part of a global event that brought together more than 1,200 people, both virtually and in person. For two full days, the participants gathered at the main campus of Humboldt-Universität zu Berlin to find new ways of accelerating discoveries in materials and chemistry with artificial intelligence (AI).

This year’s projects highlighted how AI is moving beyond simple text generation to become a true partner in science. The tools are now helping researchers plan experiments, test ideas, and even catch mistakes. They can guide entire processes – from setting up a computer simulation to improving a design step by step. Some projects showed how combining information from many sources and linking it together makes the systems smarter and more adaptable to scientific challenges.

FAIRmat not only hosted the hackathon but was also represented by three teams: Parse Patrol, Ragalicious, and MatterMinds. Out of 121 submissions worldwide, Parse Patrol and Ragalicious received Visionary Awards for their outstanding ideas and were invited to present in a special showcase session.

The success of the Berlin site was also recognized internationally. Science Magazine featured it in an article by Zack Savitsky, Researchers customize AI tools at global ‘hackathon’, which highlights how LLMs are being customized for scientific discovery, the importance of interdisciplinary collaboration, and the crucial role of high-quality research data.

We are pleased to share a recent study from Friedrich-Schiller-Universität Jena in collaboration with international partners, led by Dr. Kevin Maik Jablonka, FAIRmat AI Toolkit task leader. The study explores the capabilities and limitations of multimodal AI models in chemistry and materials science, using the newly developed evaluation framework MaCBench.

Published in Nature Computational Science under the title "Probing the limitations of multimodal language models for chemistry and materials research", the study is the first to provide a systematic assessment of how current vision language models process visual and textual information in a scientific context.

MaCBench provides an extensive benchmark of over 1,100 realistic tasks across three key areas of scientific work: extracting data from literature, understanding laboratory and simulation experiments, and interpreting measurement results. Using this benchmark, the team demonstrated that while vision language models excel at tasks such as recognizing laboratory equipment or extracting standardized data, they struggle with complex scientific reasoning, linking information from different sources, and processing visual information. The study also revealed a strong correlation between model performance and the frequency of test materials in online sources. This suggests that models may rely more on pattern recognition from training data than on genuine scientific understanding.

This research addresses a previously unresolved issue in AI research: how to fairly evaluate multimodal systems when it is uncertain which data the models have encountered during training. This advance lays important groundwork for developing future AI assistants that can reliably support scientific workflows.

The 7th edition of the FAIRmat newsletter is now available for download! Discover the latest project developments, explore new NOMAD tools, read an interview with our domain expert Esma Birsen Boydas, and find more exciting articles from the FAIRmat community. Download it now from our website!

We are excited to share the recordings of the talks from our latest Users Meeting! The full playlist includes:

• NOMAD: A Federated Research Data Infrastructure – Hampus Näsström
• Quantitative XPS Analysis Using Convolutional Neural Networks – Lukas Pielsticker
• Small Data Models for Materials Design – Luca Ghiringhelli
• The Alexandria Database of Materials – Miguel Marques
• Creating Data Analysis Pipelines Using the MADAS Framework – Martin Kuban
• Artificial Neural Network Inference on FPGAs in PUNCH4NFDI – Johann C. Voigt
• Towards FAIR Workflows in Computational Materials Science – Sarath Menon
• Neuro-symbolic Organization of Research Contributions with Knowledge Graphs and Large Language Models – Sören Auer
• Text Mining in Materials Science – Markus Stricker
• Transforming Materials Science with Transformers – Nawaf Alampara
• Large Language Models for Scientific Data Extraction – Mara Schilling-Wilhelmi

Subscribe to the FAIRmat and NOMAD YouTube channel now to stay updated on our latest insightful videos!

Janine George’s seminar from June 5, 2025 is available to watch!

In her talk, "Materials Design Using Chemical Heuristics, Workflows, and Machine Learning", she explores how automated workflows and analysis tools enable large-scale quantum-chemical bonding analysis and support the development of machine-learning models and intuitive understandings of material properties. 

Watch the full recording on our YouTube channel.

On July 2-3, 2025 the FAIRmat community gathered for the sixth time to celebrate research data management (RDM) and its role in accelerating scientific discovery. This time, the focus was on AI- and ML-driven materials science and how the FAIR data principles make research data "AI-ready". FAIRmat domain experts and NOMAD users shared their insights and experiences on using NOMAD in next-generation materials discovery.

The morning sessions featured ten talks across four key topics:

• Harnessing Data-Driven Methods in Materials Discovery
• Exploring AI & ML in Simulations, Databases, and Computational Workflows
• AI & ML Activities Across NFDI Consortia
• Language Models for Materials Science

Recordings of the talks will be available soon on our YouTube channel. 

As always, hands-on sessions were an integral part of the Users Meeting. This time, the participants engaged with the NOMAD ecosystem through two workshops:

• Role-Based NOMAD Usage and Development led by Joseph Rudzinski and Hampus Näsström
• Large Language Models for Scientific Data Extraction led by Mara Schilling-Wilhelmi and Sharat Patil

A highlight of the event was the direct and informal interaction with NOMAD developers and FAIRmat experts. Participants could get support for their specific challenges – even faster than on our Discord channel! 

The event would not be complete without a poster sessions. We thank all participants for their contributions and for their openness in strengthening the growing community of researchers advancing FAIR data practices and AI-driven materials science. You can find the full list of contributions on the website. 

We are grateful to the Bochum team for their kind invitation and warm hospitality at Ruhr-Universität Bochum.

Stay tuned for updates on the next Users Meeting later this year! 

Once again, our team took part in the wonderful Long Night of the Sciences on June 28, 2025. We had the opportunity to engage with the public, present our ideas, and witness the genuine enthusiasm and curiosity of our visitors – an experience that was both inspiring and rewarding.

It was a pleasure to see the excitement of those who stopped by our booth and shared their creative responses to the research data management task and algorithm coding challenge.

We are grateful to our colleagues at CSMB for their excellent organization and support throughout the event.

We are looking forward to seeing you again next year!