FAIR digital objects, persistent identifiers and machine actionability

Persistent identifiers (PID) and associated metadata are essential components of FAIR implementation [16]. By uniquely identifying digital objects and metadata, PID systems, services and associated policies ensure content accessibility and resolution. Therefore, providing the foundation for findability and accessibility – important aspects of FAIR Principle F1 that states both data and metadata must have a globally unique and persistent identifier [23]. Over the past several years, initiatives such as the PID Kernel Working Group of the Research Data Alliance (RDA) [18] and the Freya project [2] have identified and outlined the benefit of metadata in the context of data findability and linking. Some of these discussions have been continued and evolved within the conceptual framework of FAIR Digital Object [7]. The FAIR Digital Objects Forum [12] is now facilitating further specification work by looking into PID attributes and specific aspects of machine readability, interpretability, and actionability. The reason for these three distinctions is that structured content (such as JSON, JSON-LD, RDF, XML) is essential for machine readability but the machine also requires semantic artefacts to interpret the nature of the digital object (such as object type and relationship with other entities). And actionability here means machines can perform automated processing of digital objects [20].

The metadata elements linked to the PID record must enable operations at a higher abstraction level because FAIR Digital Objects are domain agnostic. The clear distinction between data and metadata, as outlined in FAIR principles F1 and F3, is adhered to by this higher abstraction. The abstraction also permits a further degree of separation between metadata at the PID record level and that at the content/bitstream level. The PID record may include information such as the type of digital object (whether it is a dataset or a scholarly article), a list of permitted actions (read, write, and update), version data, and details about license and usage. The content level metadata for instance can contain other elements like a document’s author, the file size, and image metadata, details about the variables in the datasets. While the PID level metadata can specify common components (either at the domain or specific use case level), the content level metadata can differ among digital objects. In other words, the PID level metadata provides a generic mechanism to describe different types of digital objects prior to accessing or resolving the content. The PID Kernel Working group defined the collection of these PID metadata as the PID Kernel Information [18]. Based on the RDA working group’s output, a working draft from the FDO forum is proposing the term FDO Record to highlight that there could be possible implementation of FDO without explicitly relying on the attributes stored in a PID record [12]. DOI kernel [10] from the DOI Foundation and DataCite Metadata Kernel [5] also provide similar minimum metadata which are records needed to describe the digital object. These PID level metadata can be implemented in the PID system or a separate database if the PID system does not allow the functionality of adding extra metadata during the PID creation and registration. This article highlights the importance of PID records and PID level metadata for FAIR implementation and recommends adding PID records for all digital object types.

Along with the precise object definition, semantics [1], vocabularies [3], and provenance [13], different types of digital objects (e.g., data, software, concepts, algorithms, workflows etc.) and metadata are playing a more and more significant role in data intensive research. These building blocks are important for data and information to be “Fully AI-ready” (another interpretation of the FAIR acronym) where the concept of machine actionability is central. A 2022 paper entitled “FAIR Digital Twins for Data-Intensive Research” by Schultes et al. [19] proposes a modular approach to building systems based on FAIR Digital Objects where best practices adoption around identifiers and associated metadata will help.

Fig. 1.

Handle PID record: https://hdl.handle.net/10.1038/sdata.2016.18?noredirect of the FAIR principle article. These metadata can also be harvested via an API.

There are currently several types of PID implementation available, including DOI, Handle, ARK, and several registration agencies that allow the use of various metadata schema. For example, in Fig. 1, the PID record of the 2016 FAIR principle journal article [23] shows only the URL and the timestamp rather than details about the type and nature of this digital object. Machine actionability is here limited due to a lack of information and specification, which prevents machines from inferring possible actions. Further information can be found via the Crossref API record [4] as Crossref was the DOI registration agency for this object. However, the DataCite API [6] returns a 404 error when someone searches for this article in DataCite. Additionally, the PID record in the Handle server (Fig. 1) does not provide the information that Crossref was used to register the DOI. Users can undoubtedly get this material from a multitude of sources. The point is that because programmatic operation support is currently dispersed, machine actionable operations are difficult to create. Adopting minimal metadata at the PID level, which is based on the FAIR Digital Object framework, is one option to improve this.

Below, I provide three examples of PID records from three different domains that show a few common elements and possibilities of different operations (two examples are borrowed from [21]).

1.DiSSCo PID record (Fig. 2)

The Distributed System of Scientific Collections (DiSSCo [9]) is a new Research Infrastructure (RI) for Natural Science Collections that uses the FAIR Digital Object concept as the core element for the data architecture [15]. In this example, the digitalObjectType states that this is a “Digital Specimen”. The object types and type definitions can be also stored as a FAIR Digital Object in a data type registry (see example [11]). Besides the type, the other metadata elements in this example describe the version of the PID record and the status of the record – as PID lifecycle information is important to understand the current status of the object. These elements are still under discussion as DiSSCo needs to accommodate different types of specimen records and metadata profiles (from zoology and botany to rocks and fossils). There are also similar discussions going on with the IGSN implementation [22] that deals with geological and earth science samples.

Fig. 2.

DiSSCo Digital Specimen https://hdl.handle.net/20.5000.1025/KMX-F7Q-6UX?noredirect.

2.Persistent identification of instruments (Fig. 3)

The Persistent Identification of Instrument (PIDINST) schema came out of the RDA working group discussion with an interdisciplinary group. This schema provides a mechanism to register PID level records and it is PID system independent [17]. The example record shows specific information about a sensor. Each index also points to a Handle record (21.T11148/709a23220f2c3d64d1e1, [14]) in a data type registry. The goal is to describe instrument instances, including relations to other entities such as the model or manufacturing details in a more standard way as instruments are essential for different types of research activities. These metadata are important for reusability, reproducibility, and experiment validation. PID level records can help different workflows to figure out what the instrument is for instance or what measurements were taken in what form.

Fig. 3.

PID for instruments: https://hdl.handle.net/21.T11998/0000-001A-3905-1?noredirect.

Fig. 4.

DARIAH dataset: https://hdl.handle.net/21.11113/0000-000B-CA4C-D?noredirect.