How are we standardising metadata in the Physical Sciences Data Infrastructure (PSDI)?

How are we standardising metadata in the Physical Sciences Data Infrastructure (PSDI)? Dr Aileen Day (PSDI metadata lead) 4th Ontologies4Chem workshop (11th November 2025) Outline Overview of current PSDI metadata Onboarding a PSDI resource Future development: Areas for formal definition Outline Overview of current PSDI metadata Onboarding a PSDI resource Future development: Areas for formal definition What is PSDI? UK funded project through the UKRI Digital Research Infrastructure theme (DRI) via EPSRC Developed out of a community statement of need Lead institutions: University of Southampton and Science and Technology Facilities Council (STFC) First launch of resources are now live! https://www.psdi.ac.uk PSDI: filling a Gap in Provision Other domains have established provision in the UK, e.g. EBI in Life Sciences NERC Data centres in Environmental Science UK Data Archive in Social Science USA: Materials Genome Initiative Japan: NIMS European data infrastructures, e.g. E-CAM, MaX and NOMAD German National Research Data Infrastructure (NFDI) Other countries have initiatives underway in this domain, e.g. We are building a UK, Physical Science, Data Infrastructure Supporting UK Chemistry, Materials and related disciplines Interfacing to other domains: eg. Life, Medical, Engineering and Environmental Sciences Interfacing with international initiatives the glue that brings PSDI together PSDI metadata https://data-search.psdi.ac.uk/ https://resources.psdi.ac.uk/ PSDI Metadata Top-level PSDI metadata Documentation PSDI Metadata – top-down approach Metadata about resources Metadata about properties within resources Metadata about data within resources Metadata about data provenance Metadata that supports PSDI Resource Catalogue Metadata that supports PSDI Cross Data Search PSDI Metadata – using existing standards Wherever possible we use existing standards, ontologies, vocabularies, community best practices e.g. Following CDIF1,2 recommendations CDIF is “first and foremost a practical exercise: how can we find a set of common standards and technology approaches that will enable us to implement FAIR for cross-domain scenarios using things which exist today” 1. D2 .3 Cross-Domain Interoperability Framework CDIF Report (Synthesising Recommendations for Disciplines and Cross-Disciplinary Research Area report) is a recent WorldFAIR output 2. https://cross-domain-interoperability-framework.github.io/cdifbook/introduction.html PSDI Metadata DCAT Resource Catalogue What we Provide - PSDI (https://resources.psdi.ac.uk/ ) PSDI Metadata DCAT Resource Catalogue Resources, descriptions and details are retrieved from DCAT jsonld metadata https://metadata.psdi.ac.uk/psdi-dcat.jsonld DCAT format (https://www.w3.org/TR/vocab-dcat-3/ ) Onboarding a PSDI resource step 2: resource description for catalogue Python scripts feed these into psdidcat.jsonld Currently gather resource theme and resource descriptions in Excel templates, collaboratively with resource owners Onboarding a PSDI resource Step 3 PSDI PSDI terminology PSDI Resource Catalogue PSDI Cross Data Search SKOS vocabulary DCAT resource catalogue OPTIMADE PSDI Cross Data Search PSDI Cross Data Search uses OPTIMADE We require the resource data to align with OPTIMADE core properties and PSDI namespace properties for maximum searchability… PSDI Cross Data Search Align with OPTIMADE properties in: -structures endpoint (https://schemas.optimade.org/defs/v1 .2/entrytypes/optimade/structures ) -references endpoint (https://schemas.optimade.org/defs/v1 .2/entrytypes/optimade/references ) PSDI – find a substance What is a substance? everything made up of elements searching on OPTIMADE structure endpoint property: descriptive chemical formula (chemical_formula_descriptive) This is the most widely-populated chemical formula property but others are available for searching in the “Advanced Search”: reduced chemical formula (chemical_formula_reduced) Hill chemical formula (chemical_formula_hill) anonymous chemical formula (chemical_formula_anonymous) OPTIMADE provides comprehensive definition of core properties e.g. PSDI – find a substance searching on OPTIMADE structure endpoint property: number of elements (nelements) elements elements ratios (elements_ratios) What is a substance? everything made up of elements If it is a crystal structure searching on custom properties for cell parameters: a, b, c, alpha, beta, gamma will be linked up with definitions in the cif namespace (note that OPTIMADE captures the alternative lattice vectors (lattice_vectors) which we do not use but can be searched on in Advanced Search) PSDI – find a substance Outline Overview of current PSDI metadata Onboarding a PSDI resource Future development: Areas for formal definition Resource subject classification (in progress) Substance definition Units Source of funding/project identifiers Resource subject categorisation Aim: review choice of dcat:themeTaxonomy (taxonomy/vocabulary/ontology/classification system) to describe the dcat:theme (research topic or scientific discipline classification) of PSDI resource themes and resources. This is used: 1. In the DCAT metadata which describes the catalogue of resources and resource themes 2. In the PSDI What We Provide resource catalogue to facilitate resource filtering 3. In the PSDI Cross Data Search to facilitate data source filtering Currently using EuroSciVoc but this decision was taken quickly and does not include implementation of hierarchy Note that while DCAT calls this field dcat:theme, we will refer to it explicitly as dcat:theme or “subject” here (rather than just “theme”) to avoid confusion with resource theme Current resource subject categorisation 1. In the DCAT metadata which describes the catalogue of resources and resource themes https://metadata.psdi.ac.uk/psdi-dcat.jsonld Current resource subject categorisation 2. In the PSDI What We Provide PSDI resource catalogue to facilitate resource filtering (currently the terms themselves are shown with no hierarchy) Resource subject categorisation requirements Must have: 1. Intuitive: straightforward for resource-owners and users to understand and navigate 2. Physical Sciences scope: includes the whole of the physical sciences, not just part of it 3. Maintained and versioned: updated as research evolves (so that new terms are added and old terms are still preserved) Nice to have: 1. Persistent identifiers: for terms (we could formalise an existing classification system into a skos vocabulary with persistent identifiers if it does not currently have this format) 2. Definitions: term definitions to make it clearer to navigate and understand 3. Hierarchical: so that it can provide different levels of detail as required Subject classification selection process Review different subject classification systems according to requirements Test shortlisted candidates 1. How well do they accommodate the PSDI resources as of summer 2025? Description of review will shortly be published on Zenodo in the PSDI community (https://zenodo.org/communities/psdi/records?q=&l=list&p=1&s=10&sort=newest ) Resource subject categorisation candidates investigated Nature Physical Sciences categories FORD (Fields of Research and Development)/FOS (Fields of Study) from Frascati Manual Resource subject categorisation candidates investigated Also useful to see various ways of organising subject and good ways of doing things: organise consistently by: Discipline (e.g. chemistry) Sub-discipline (e.g. organic chemistry) Specific field Include “multidisciplinary” and “other” at top level Shortlisted candidates Classification system Pros Cons EuroSciVoc - Top level is well established (FORD/FOS), granularity added by NLP - Used by European Union to classify and search on research projects in CORDIS - We are already using it in PSDI - Semantic with pids - Top-level is a bit confusing, so would omit in PSDI - Confusing definition of “physical sciences” in PSDI context - No definitions The Modern Science Ontology ( modsci) - Used by SciData which is one option for PSDI to consider for data-level metadata - Very semantic (BFO ontology) with pids - Some definitions - Ontology does not just describe scientific disciplines, but also discoveries, phenomena, scientists, instruments - Top level “Applied Sciences”/”Formal Science”/”Interdisciplinary Science”/“Natural Sciences” is maybe not so would omit in PSDI - Good representation of Chemistry but can be patchy coverage elsewhere e.g. “Computer Science” OpenAlex Topics - was Fields of study taxonomy within Microsoft Academic Graph (MAG) - Now used to categorise OpenAlex - Top level domains, fields and subfields correspond to Elsevier’s Scopus’s All Science Journal Classifications (ASJC) structure are very intuitive and organised - There is accompanying open source code to assign topic based on title and abstract - Descriptions - Fields, subfields and topics can be linked to via OpenAlex API (e.g. https://openalex.org/topics/T10464 ) - Bottom level (topic) is arguably too detailed and specific - Bottom level (topic) is generated by an LLM and assigned to the higher level hierarchy - some inconsistencies at this level e.g. The best description for crystallography (10464 Crystal structures of chemical compounds) sits under 1604 Inorganic Chemistry rather than Organic chemistry - Would be most work to implement Would require ontology development Inconsistencies of LLM-generated bottom-level (revisit in time) Resource subject categorisation Next steps Review current EuroSciVoc term assignment to resource themes and resources with PSDI project partners and update psdi-dcat.jsonld accordingly Add mapping of dcat:theme values to their containing folders (EuroSciVoc subset) in metadata.psdi.ac.uk to help with this Adapt “PSDI What We Provide” resource catalogue filters to show hierarchical categories for subject Add the same filters to PSDI Cross Data search when selecting Data Sources to search Publish PSDI_Subject_Requirements.docx on Zenodo and link it to DCAT metadata documentation