Providing a chain of provenance for specimens and related data is a major challenge and has a set of roadblocks along multiple dimensions. Traditional field collecting methods are ingrained in many scientists. The informatics community needs to reach out more effectively and explain to scientists the limitations of existing workflows and why an identifier scheme built around global uniqueness is not only necessary from an informatics standpoint, but would dramatically enhance the value of data for re-use, syntheses and analyses. Identifier solutions must support scientists’ current practices and create minimal burden during the collecting process. The solution should provide incentives for adoption, both in the field and in downstream information systems. In particular, effort is needed to ensure perpetuation of field-assigned identifiers through to more permanent data curation steps. Whatever underlying identifier system is chosen, it needs to be robust in preventing the same identifiers from being assigned to different objects (and, ideally, reducing circumstances where the same object receives multiple identifiers).

An additional roadblock is a lack of clarity as to which classes of objects, concepts, or events identifiers should be assigned. Should GUIDs be associated with the actual, physical specimen or with the derived digital (e.g. images) or physical (e.g. tissues) derivatives? Focusing on biocollections specimens as material samples helps semantically clarify what bears the identifier, but many other modeling challenges relating measurement processes etc. to specimens still remain. Even for physical specimens, there are challenges in defining the types of entities that can constitute a specimen, which range from a distinct organism to a part of an organism, to a set of organisms, to abiotic samples containing specimens (e.g., a jar of seawater).

For newly collected samples, a highly desirable next step is the ability to assign globally unique identifiers directly to newly collected specimens or mixed samples in the field or shortly thereafter. In many cases, it may be desirable that these identifiers be pre-minted and written into a physical barcode or QR-Code, perhaps in conjunction with a human-friendly identifier. Figures 1 and 2 show different examples, the first representing a traditional biocollections object and the second depicting mass-labeling of tubes associated with collections samples. Assigning GUIDs to specimens at the time of collection allows field researchers to publish references to recently collected specimens without waiting for institutional identifiers that are assigned during the accession process. Beyond simply assigning unique identifiers in the field, it is critical that these identifiers persist perpetually with the objects they identify and all descendant samples, subsamples, analyses, data and publications referring to them, ensuring an unbroken chain of data provenance. In the best of all possible worlds, identifiers assigned in the field are retained as the permanent institutional identifier during accessioning.

Figure 1. 

Example of UUIDs embedded within QR-Codes on microcentrifuge tube labels. The 5 mm × 5 mm QR-Codes (Version 2) are printed with a standard laser printer on sheets of self-adhesive 9 mm dots, and scan reliably with a standard barcode reader, while still providing room for a human-readable 5-character prefix + 5-digit number (the human-readable number and UUID are permanently cross-linked in the data management system). Photo: Robert K. Whitton.

Figure 2. 

Example of a PURL-URI as a QR-Code, in this example attached to a digitised lichen type specimen in the Natural History Museum, University of Oslo. The QR-Code corresponds to http://purl.org/nhmuio/id/c1a8b878-a4f9-448b-be00-26cbad58b11c.

It is not feasible (or, at this stage, even desirable) for the entire biodiversity community to adopt a single implementation for identifiers. However, evaluation of the available technical solutions is a high priority, and the scope of solutions includes IGSNs, DOIs, EZID ARKs, LOD-URIs and UUIDs (comparisons among many of the different options are shown in Table 2 and a comparison of more or less centrally managed mapping and redirection services is shown in Figure 3). The group explored several different viewpoints promoting the utilization of HTTP URIs for all identifiers and did not reach a consensus. HTTP URIs have the advantage that they provide a semantic web compatible default dereferencing method through the standard http protocol and can be flexibly constructed (Hagedorn et al. 2013). The advantage of many identifiers not being a HTTP URI is that the omission of a default dereferencing method avoids potential confusion and may allow for even greater flexibility. However, we recommend all identifiers have the ability to be dereferenceable through at least one http-based service, even if the http-form is not preferred.

Figure 3. 

Identifier schemes differ in whether redirections and mappings to ensure stability are centrally managed or not. Top: a DOI dereferencing service like CrossRef or Datacite redirects to the actual content provider; the URIs of content data and RDF metadata are publicly visible and can be used as independent (albeit often unstable) identifiers. Bottom: A linked open data pattern, where each content provider assumes the responsibility for maintaining a stable mapping; the content negotiation is internal. Modified after Hagedorn 2013.

The group strongly suggested that an immediate next step would be to prototype solutions to create persistent identifiers built on different, existing platforms. Such prototypes would engage stakeholders in testing and feedback in order to refine prototypes. The prototypes could also spawn key actions, including more focused workshops/hackathons, perhaps in the context of the Taxonomic Databases Working Group meetings (TDWG), with the goal of reporting outputs of such trials. TDWG, in particular, is a crucial stakeholder as an international standards organization for biodiversity objects and data.

Scaling up to a larger system will require obtaining funding to support development. A fruitful path would be to align a few organizations that are working nationally or globally (e.g., NEON, iPlant (http://iplant.org), iDigBio, Critical Zone Observatories, Consortium of European Taxonomic Facilities) to adopt an early version of the system and to show interoperability and enhanced ability for tracking specimens and their derivatives as an outcome. For those more at the longer-tail of the specimen curation process, such as smaller biocollections or individual labs, incentives for adopting a system to replace the local numbering systems currently in practice could help coalesce efforts, and could further promote the value of such approaches when putting together data management plans (DMP) for funding agencies. In particular, identifier-specific DMP Tool (https://dmptool.org/) template content should be provided. Finally, with the strong growth of handheld devices, the biodiversity informatics community should work to produce tools for assigning identifiers with such devices.

A more detailed implementation proposal could be specified just for field collections, as part of a TDWG task group, leading to a community input and review process. This would be one key part of a larger effort to identify and reach out to national and international stakeholder groups, including collection managers, aggregators, publishers, scientists, funding agencies, downstream users of the data, and developers of software (e.g., Specify, http://specifyx.specifysoftware.org/; Symbiota, http://symbiota.org/docs/; and in-house software used by aggregators such as GBIF).

The single key roadblock with legacy data is the use of local identifiers at all steps during the collection and accessioning process. While these provide means for local provenance tracking, they are insufficient for managing across collections, and are hard to adapt and scale to an open platform for data discovery such as the Internet. A classic example is botanical “duplicates” that come from the same collecting event where different clippings of the same plant were sent to multiple museums. Similar issues can be found for cases where specimens were gifted from one collection to another. In these cases, linkages associating those specimens across collections were typically severed when biocollections were accessioned independently into institutional museum repositories. Those past associations can only be inferred from re-compiling data and looking for content-level matches related to the collections events.

Further, because most collections have effectively developed local curatorial practices, often based on regional and taxon-specific approaches, there is a wide variety of different legacy identifiers associated with specimens and their data. In sum, current practices were and remain highly heterogeneous and the information that could re-associate specimens across collections are lost and cannot be solved simply via post-hoc application of new GUIDs. Thus, the problems with legacy collections are managing both identifiers already in use and dealing with potential application of new ones.

Next Steps: As a pragmatic matter, the immediate next steps for legacy collections may not include broad application of globally unique identifiers. Instead, a short term next step is for biodiversity informaticians and collections staff to work together to standardize practices for assigning unique identifiers that are persistent (remain tightly associated with the objects they identify) and stable (continue to be actionable). At a minimum, institutions should clarify the identifier scheme being used locally via their own internal policies. Further development of community-wide best practices would be more effective because they would not only foster local curatorial practice, but also specify how those locally curated materials and their data eventually become part of the rapidly coalescing global, digital framework. These best practices need to be developed in the context of existing efforts and/or organizations such as the Global Registry of Biorepositories (GRBio; http://grbio.org/), which provides a needed framework for publishing repository-specific information like standard acronyms for institutions and collections. Curators should register their collections in GRBio and specify the adopted identifier scheme for the collection.

The legacy group also considered medium-term and longer-term goals, focusing more on broad informatics solutions than local identifier curation practices. One critical step is to assemble identifiers published by curators to aggregators such as GBIF and to assess identifier heterogeneity. This can feed into developing software for comparing identifiers (e.g., resolvers) that is better able to perform fuzzy matching on identifier strings (and fetch such variations), given that identifiers are sometimes expressed in unintended ways (e.g., added spaces or hyphens, capitalization, etc.). Using just a simple string comparison is insufficient and more robust systems should be set in place, which will then forward to the correct identifier. The same applies to whether a URI prefix should be part of the identifier or not.

Next, in order to avoid broken URIs, institution-independent resolvers (e.g., purl.org) or aggregators (e.g., GBIF) should check dereferenceable URIs at certain intervals and inform the responsible contact person when the target URIs return a 404 HTTP status code or are otherwise unavailable. Some providers, such as CrossRef (http://www.crossref.org/), offer services for policing broken URIs. With regards to the data records associated with specimens and published to aggregators such as GBIF, the legacy identifiers group strongly argued for the longer-term goal of inclusion of proper GUIDs in the occurrenceID (or materialSampleID) Darwin Core (Wieczorek et al. 2012) field, rather than some sort of concatenation of local identifiers, such as a Darwin Core triplet (Guralnick et al. 2014). Finally, we strongly encourage integration of identifier metadata into existing standard schemas (e.g., Darwin Core, ABCD; http://www.tdwg.org/activities/abcd/) as new concepts. Such metadata would include information regarding identifiers, persistence, rules for attribution (use, citation, reference) etc.. as is also discussed further in the “cross-cutting solutions” section.

The legacy biocollections group developed a list of immediate action items to most efficiently take the steps listed above. As a priority list, these include:

1. Assemble current identifiers from aggregate data as a means to determine current practices. Some of this work has already been accomplished as part of work by Guralnick et al. (2014) to evaluate Darwin Core Triplets and their current use as identifiers in different systems (e.g., VertNet, http://vertnet.org; Barcode of Life Data Systems, http://www.boldsystems.org/; GenBank), but further work focusing on GBIF datasets is needed. A critical assessment of current implementations will feed into the next step of generating more informed best practices or appropriate strategies that individual institutions can adopt based on their current GUIDs application.

2. Create best practice documentation on known identifier minting schemes. Document best practices with use cases, examples, and pros and cons.

3. As in the new field-collected biocollections group, there is a need to further clarify what exactly is being identified - MaterialSample vs. Organism vs. Occurrence; physical object vs. digital representation.

4. Clearly define the scope of the proposed identifying scheme and what benefits can be gained by it.

5. Demonstrate the implications for publishing in the primary literature.

The difficulties in managing, tracking, and large-scale extracting of citations from any sources other than traditional publications are, in part, due to the paucity of widely adopted, persistent, globally unique and resolvable biodiversity data identifiers. Additionally, extracting specimen, taxon, and other biodiversity data from modern scholarly publications with unstructured formats and little to no markup is needlessly challenging. Another major obstacle is that information about specimens might be published in different places and with different levels of granularity. For example, a specimen might be cited as a holotype in a protologue, then georeferenced and published again in subsequent revisions, perhaps even under a synonym, with images and DNA data appearing separately in other publications. Unless the original specimen collection number is used consistently across all publications, it is difficult, if not impossible, to link together all the important digital derivatives independently generated from that specimen.

A final roadblock is the lack of adoption of advanced publishing approaches, including semantic markup, by almost all publishers in this domain. The TaxPub /Journal Archival Tag Suite provides (Catapano 2010, Penev et al. 2010, 2012) all the necessary functionality and has been successfully implemented by Pensoft in 14 journals, including the registration of the their articles in PubMed and PubMedCentral. However, it places the burden on publishers to adopt new technical approaches that are difficult to meet given a lack of resources and strong incentives for change.

The key next step is to establish the best practices to generate and assign identifiers as they either propagate from biocollections into the literature or are created during semantically enabled publishing processes. Such practices will assure that publications follow a set of principles ratified by various stakeholders and governments, and perhaps best described broadly in the Force11 data citation principles (https://www.force11.org/datacitation), and more directly for the biodiversity community in the Bouchout Declaration (Bouchout Declaration 2014, http://bouchoutdeclaration.org/). Tools are needed to retrieve identifiers assigned to biological names, taxonomic treatments associated with a name and specimen data discovered in the published records and/or stored in domain specific databases.

Below we summarize critical practices and principles for the use of identifiers in semantically enhanced publications:

1. Publishers should use GUIDs for formally cited or potentially relevant data (e.g., authors, books, articles, taxon names, taxonomic treatments, gene sequences, specimens, etc.) maintained in well- established and widely used external registries.

2. Publishers should issue GUIDs for data first made widely available through document publication (e.g., observation on a species published by an amateur naturalist with no GUID issued by or associated with an Institution).

3. Publishers should provide both human- and machine-readable content (Starr et al. 2015) through resolvable GUIDs for separate elements of an article (e.g., individual images, graphs, tables, supplementary materials, taxonomic treatments, checklists, etc.).

4. Resolvable GUIDs should be used as widely as possible to annotate published content; for example, adding a species to a published checklist should be identified by a GUID, which can be linked to the exact “place” within a published text (e.g., between two species in the checklist).

5. Publishers should use GUIDs and authority files for authors, e.g., ORCID (http://orcid.org/), VIAF (http://viaf.org/), authors of plant names (http://www.kew.org/data/authors.html), ZooBank authors (http://zoobank.org) or internal systems that unambiguously identify names of authors.

6. For the conversion of legacy literature, assign GUIDs to relevant elements that are widely used, resolve to content (e.g., articles, treatments, observation records) and can be a source for Linked Open Data. Whenever possible, use an existing identifier service (such as Plazi for treatments), rather than minting additional identifiers.

7. The identifier system(s) should be sustainable for the long term, highly reliable, and have an API as a backbone service.

8. We note a preference for identifiers used by indexing services (while such services use many kinds of identifiers, CrossRef and DataCite (http://datacite.org) DOIs are the most commonly used). Publishers should link data related to an article and the article itself through their GUIDs (CrossRef and DataCite DOIs cross-referencing service).

9. Identifiers and their metadata related to annotations in publications should be housed and made available by an independent party.

We discussed how systems can be built around identifiers that support all the different participants involved in publishing. Authors are critical participants and should better be able to cite usage of their data from semantically enhanced, rather than unstructured, formats. Publishers can assist authors by making all published data linkable/citable and contributing to specialized databases and/or permanent repositories (e.g. Dryad (http://datadryad.org/) or the Biodiversity Literature Repository (https://zenodo.org/collection/user-biosyslit)). Publishers can also provide authoring tools (such as the Pensoft Writing Tool (PWT) used by the Biodiversity Data Journal – see Smith et al. 2013) that assist authors with entry of structured data (i.e., upfront pre-submission markup and easy data import into the manuscript) to which new or existing identifiers can be assigned or included. Hence, easy data download and export to aggregators from the published paper can be achieved.

To serve the broader community (i.e., beyond authors), publishers can also provide tools to find cited data (e.g., http://refindit.org, which searches across CrossRef, DataCite, Mendeley (http://www.mendeley.com), RefBank (http://refbank.org), Global Names Usage Bank (http://www.globalnames.org/GNUB), Biodiversity Heritage Library (http://www.biodiversitylibrary.org/), Biodiversity Literature Repository (https://zenodo.org/collection/user-biosyslit) and others), as well as an ORCID lookup linked to data creators or owners. Contributing institutions can much more easily assess their institutional impact in biodiversity research output by tracking the usage of identifiers embedded in the articles, as well as better manage intellectual property. For example, publishers can work with organizations such as GRBio to create identifiers for institutions so that all can be cited. Funding agencies can better argue that open access is not only a legal mandate but maximizes their return on investments in terms of products made available to the public. One possible step forward is to create identifiers for funding agencies (e.g., Fundref http://www.crossref.org/fundref/).

Not all identifier schemes are unambiguous in declaring which identifier refers to an information resource and which to a physical object or abstract concept or event. For instance, an identifier referencing a photo of an eagle on a tree could be identifying the digital photo itself, a photographic print that was later scanned, a reference to the eagle as a physical specimen stored in a museum, the event of capturing the image, or a reference to an individual eagle that exists in nature. Distinguishing concepts such as “digital media”, “print media”, “individual”, and “specimen” is not trivial and ultimately relies on attaching formal descriptions from a biodiversity or biocollections ontology to the identified object. We encourage the use of the Darwin Core Basis Of Record term (http://rs.tdwg.org/dwc/terms/basisOfRecord) to describe the exact nature of the resource. There is a current proposal for tying values for the Basis Of Record term to ontology sources in the Biological Collections Ontology (Walls et al. 2014, Deck et al. in press) which will greatly help in clarifying the concepts underlying identified objects and their downstream use.

Various identifier schemes behave differently when posting requests and receiving responses; standardized responses are urgently needed. An important example is the standardized content negotiation behavior in the semantic web; other examples are the unified content negotiation by CrossRef and DataCite (http://crosstech.crossref.org/2012/05/crossref_and_datacite_unify_su.html). Identifier metadata can be requested from the service provider not only using Linked Data patterns (which a user cannot do with just a web browser), but also by manipulating the URL endpoint directly, such as URL inflections (https://wiki.ucop.edu/display/Curation/ARK), alternate resolution prefixes, 303 re-directs or hashtags to denote physical objects, or parameter specification in the URL query string. The EZID system provides the ability to deliver DataCite, Dublin Core, CrossRef, or Dublin Core kernel (http://dublincore.org/groups/kernel/spec/) metadata profiles. A strong recommendation is to create a biodiversity metadata profile to complement these existing profiles.

What intention goes into the creation of an identifier, including any contracts and technical specifications? The policies of identifier assigning authority provide information about the expectation of commitment, longevity, use, and re-use. Some identifier schemes require membership and fees in order to create identifiers while others are open and free. Some schemes mandate use of a particular table lookup technology while others do not. Each scheme has its unique history, community, and conditions of use (as described in more detail in Table 2). Whatever method is used for creating the identifier, it should be publicized explicitly by the identifier authority. Consumers need to know about the persistence mission of the agency and any potential contracts implied by use of the identifier.

The group discussed issues with contracts about retaining identifiers in downstream systems. We strongly recommend creating community conventions when re-using data to place special importance on referencing and maintaining earlier identifiers, especially those with clear policy and behaviour contracts. Use of such conventions provides significant value for data producers and consumers, such as data citation networks, analogous to those produced by CrossRef for journal publications.

If a physical object is categorized as “organism” and is later changed to “bulk sample”, does its associated identifier change with it? Does the identifier to a concept change if a spelling error or ambiguous wording is corrected in its definition? Does an information resource identifier change during versioning? Does an identifier guarantee binary identical results, or only identical core-content (which may be embedded in a modified template or formatted differently)? In some cases the answer may be a permanent single mutability policy of the identifier scheme itself, in other cases the identifier scheme may support multiple policies, and the mutability policy may be available as metadata on the identified object. We recommend using, and where necessary, developing, a vocabulary to document mutability policies and conventions for various content types.

Dereferencing, or the automated process that a software tool (e.g., a web browser) employs to go from identifier to content or metadata access, starts with a URL. All identifiers, regardless of scheme, are resolved by a user agent if they are embedded in a URL. As for institutions that have long-term access in their mission, many people think that smaller, newer institutions’ website hostnames will be short-lived compared to those of older, larger institutions (e.g., loc.gov, bnf.fr). Some people prefer to trust a hostname backed by a group of institutions, even if comparatively young (e.g., dx.doi.org), rather than by any one institution. Among such group or consortial arrangements, some people prefer to trust those committed to open access (e.g., gbif.org). Persistence missions can also far exceed current technological solutions. Will, for example, current http protocols look anything like the protocols used in 2065? Forecasting about resolver persistence for 10, 20, 50, or 100 years is at best educated guesswork, but it should take into account such things as inevitable technological advances, resolver organization’s mission, size, business model, openness, and current age.

Identifier readability and ease of transcription are concerns whenever identifiers are routinely recognized, typed, or written by human beings (e.g., on specimen labels). Non-opaque identifiers (containing recognizable strings) tend to be easy to read and to enter because humans can often spot transcription errors; however, it is difficult to mint them uniquely and quickly, and to keep them persistent (their structure makes them prone to “semantic rot”). It is easy to create UUIDs quickly and in large number, which can be especially useful for tracking instances of samples or events in aggregator databases. On the other hand, UUIDs rendered as hexadecimal characters (as opposed to embedded in QR-Codes) are opaque and long, and not as useful in situations where a UUID is expected to be printed onto an insect pin, placed in a vial, or entered via a user interface by hand. There are other means of generating shorter unique opaque identifiers (e.g., Noid), but they have other disadvantages. One solution to this dilemma is to maintain human-friendly identifiers (e.g., catalog numbers) when presenting content to humans in addition to computer-friendly identifiers (LOD, UUID, DOI, ARK, etc.) for electronic cross-linking. Such a solution does require curation overhead to assure that both are managed for the long-term. Emerging services such as GRBio maps human-friendly Institution and Collection Codes to URIs for biocollections.