FDO Profiles - Dataset Attributes

[juliancervos]

As part of the FDO efforts in BioDT, we need machine-actionable profiles for all the different digital objects we use. Datasets are one of the fundamental digital objects that is widely used by different pDTs. It is this diversity of dataset types and formats that we need to pay special attention in order to accommodate them properly into the FDO framework.

The metadata attributes shown in the table below are based on the RO-Crate guidelines, which rely on Schema.org Dataset type. However, Bioschema's Dataset profile offers a simpler selection more specific for our domain. They have also been influenced by BioDT's own data tables, such as those listed here or in each pDT page, as well as the Data interviews.

There are several other (meta)data standards that could be interesting to align with. For example:

• When necessary, more detailed metadata attributes could be included from the Data Catalog Vocabulary (DCAT) - Version 3, while Darwin Core is more specific for the biodiversity domain.
• Current (meta)data models from the Research Infrastructures of BioDT, like the GBIF Data Model or the LTER B2Share metadata attributes.
• The ODMAP protocol for reporting Species Distribution Models (following the initiave for the Model Attributes FDO Profiles - Model Attributes #4 ).
• Biodiversity data cubes. This could link with the Reliance RO-Crate profile.
• HugginFace's Dataset Cards could be an interesting option for the AI-oriented datasets.

Dataset-specific Attributes Table

Table format: the name of the metadata attributes is shown in the "Property" column, including a link to Schema.org (or another schema from which the property was taken), and a definition in the "Description" column. The "Type" column specifies the expected values for each property, while the "Cardinality" column specifies the amount and whether they are optional or mandatory. Lastly, additional remarks might be added to the "Comments" column, and an example is given in the final column.

Property	Description	Type	Cardinality	Comments	Example
url	The location of a page describing the dataset.	URL	1/1		"url": "http://www.ebi.ac.uk/metabolights/MTBLS234"
variableMeasured	The variables that are measured in the dataset, either described as text or as pairs of identifier and description using PropertyValue, or more explicitly as a StatisticalVariable.	Property PropertyValue StatisticalVariable Text	1/many	Often but not necessarily each variableMeasured will have an explicit representation as (or mapping to) an property such as those defined in Schema.org, or other RDF vocabularies and "knowledge graphs". In that case the subproperty of variableMeasured called measuredProperty is applicable.	"variablesMeasured": "species cover"
measurementTechnique	A technique, method or technology used for measuring the corresponding variable(s) in the dataset. The measurementTechnique property helps when extra clarification is needed about how a measuredProperty was measured.	DefinedTerm MeasurementMethodEnum Text URL	1/many	If there are several variableMeasured properties recorded for some given data object, use a PropertyValue for each variableMeasured and attach the corresponding measurementTechnique. The value can also be from an enumeration, organized as a MeasurementMetholdEnumeration.	"measurementTechnique": "mass spectrometry"
spatialCoverage	The areas that the dataset applies to: a dataset of New York weather would have spatialCoverage which was the place: the state of New York. It is a subproperty of contentLocation.	Place	1/many		"spatialCoverage": { "@id": "https://deims.org/6ae2f712-9924-4d9c-b7e1-3ddffb30b8f1"}
temporalCoverage	The relevant time period that the dataset applies to in a precise notation, either as a DateTime or as a textual string indicating a time period in ISO 8601 time interval format.	DateTime Text URL	1/many	Open-ended date ranges can be written with ".." in place of the end date. For example, "2015-11/.." indicates a range beginning in November 2015 and with no specified final date. This is tentative and might be updated in future when ISO 8601 is officially updated.	"temporalCoverage": "2023-11-01T00:00:00Z--2023-11-13T12:10:00Z"
encodingFormat	Data type of the dataset files, typically expressed using a MIME format (see IANA site and MDN reference).	Text URL	1/many	Unregistered or niche encoding and file formats can be indicated instead via the most appropriate URL, e.g. defining Web page or a Wikipedia/Wikidata entry.	"encodingFormat": "text/csv"
contentSize	The file size of the whole dataset in bytes.	Text	1/1		"contentSize": "108MB"
version	The version number for this dataset.	Number Text	1/1		"version": "8.0.1"
publisher	The publisher of the dataset, such as the research infrastructure providing access to it.	Organization Person	0/1	In BioDT, it will typically be one of the four research infrastructures involved in the project: GBIF, eLTER, DiSSCo, or LifeWatch ERIC.	"publisher": {"@id": "https://ror.org/039kwqk96"}

Apart from the mandatory attributes above, the previously mentioned profiles and standards include many other attributes that can be used to add further details when necessary.