Skip to content Skip to footer

Your tasks: Machine actionability

What does machine-readable, machine-actionable or machine-interpretable mean for data and metadata in RDM?


More and more often, funders, data managers/stewards, IT staff and institutions in general encourage researchers in Life Sciences to generate metadata (and data) in ways that can be retrieved, read and processed by computers (machines).


  • It is common to come across different terms, such as machine-readable, machine-actionable and machine-interpretable, which express different levels of making “(meta)data for machines”. The definition and the differences between these terms are not always clear and depend on the current technology. Computers, software, programming languages, formats and standards evolve quite fast and therefore new inventions could potentially make machine-readable/actionable any digital object that wasn’t before.
    • One example is how developments in computer vision are making more and more the information contained in images, and not just the images themselves, available for processing.
  • While providing an all-encompassing definition for this topic is not within the scope of this platform, it is important to clarify that (meta)data can be used by machines to different extents, depending on its characteristics. Here, we report a few common definitions.
    • “Machine-readable: data in a data format that can be automatically read and processed by a computer, such as CSV, JSON, XML, etc. Machine-readable data must be structured data.”, Open Data Handbook.
    • “Machine-readable data, or computer-readable data, is data in a format that can be processed by a computer. Machine-readable data must be structured data.”, Wikipedia.
    • “Machine-actionable: this term refers to information that is structured in a consistent way so that machines, or computers, can be programmed against the structure.”, DDI.
    • Machine-interpretable: machines can put the provided information into context and “understand” the meaning (semantics) and relations contained in the digital object. This concept is strictly related to the Semantic Web vision and the Linked Data concept. See e.g. What Is the Semantic Web?.

    The terms machine-readable and machine-actionable are often used interchangeably as synonymous. It is because of the variety of possible definitions for data that can be processed in some form by computers, that we decided to use the term machine-actionable in the remainder of this document to refer to this type of (meta)data.

  • Machine-actionable (meta)data doesn’t mean just digital. For computers and software, it might not be possible to process the information contained in a digital object (e.g. scanned image). It is also NOT just:
    • A digital file that is readable by some software (i.e. not broken or corrupted).
    • A digital file in an open (non-proprietary) or free format (ex: .txt, .pdf) that can be read by some software.
    • A digital file that is readable by some non-proprietary software (e.g. .txt, .pdf).
  • “The appropriate machine-actionable/readable format may vary by type of data - so, for example, machine-actionable/readable formats for geographic data may differ from those for tabular data.”, Open Data Handbook. For instance, GML is one of the appropriate format for geographical information.

  • Machine-actionable/readable formats are typically difficult to read by humans. Human-readable data is “in a format that can be conveniently read by a human. Some human-readable formats, such as PDF, are not machine-actionable/readable as they are not structured data, i.e. the representation of the data on disk does not represent the actual relationships present in the data.”, Open Data Handbook.
    • For instance, have you ever tried to extract or copy-paste a table from a PDF into a spreadsheet? It is usually very difficult and sometimes even impossible. This is a practical example of why PDF is not easy to read by machines, but it is very easy to read by humans. This occurs because the content in a PDF is described as “characters painted or drawn on a space”. So text is not text and tables are not tables for a PDF. They are just characters on the page space.
    • Tabular data in CSV file can be quite easy to read by humans, unless the table is very very big. A file in CSV format can be read by machines since it is organised in records (lines) and fields (columns) separated by comma, that is as a table. So, the computer reads whatever information stored as CSV in this tabular format.


For RDM in Life Sciences, machine-actionable metadata and data should:

  • Be structured data: “data where the structural relation between elements is explicit in the way the data is stored on a computer disk.”, Open Data Handbook.
  • Be in a format that allows “many types of structure to be represtented.”, Open Data Handbook. For instance, JSON and XML for text files; certain formats for e.g. images that include structured (meta)data in a structured format.
  • Be interpreted by computer systems unambiguously. The meaning (semantic) of the (meta)data should be unique and shared among computer systems.
  • Not be in PDF format (scanned images of lab books, tables, articles or papers in .pdf)
  • Not be in plain text (.txt) nor Word documents (.docx) formats (e.g. README.txt file).
  • Not be images, audio nor video (.jpeg, png, etc).

What are the advantages of generating machine-actionable metadata and data, during and after the project?


Numerous research institutes have already introduced or are going to introduce the use of Electronic Laboratory Notebook (ELN), Laboratory Information Management System (LIMS) or similar systems to manage samples, reagents, metadata and data, during a research project. The reason for this is that this software could organize information in a structured way and make (meta)data “more” machine-actionable, compared to traditional lab books or individual folders and files in a computer. The use of machine-actionable (meta)data allows for scalable solutions that can be applied during a project’s lifetime, increasing efficiency and ensuring that findings and contributions remain relevant within the research group.

Similarly, funders and institutions ask researchers to make their (meta)data FAIR and available in a machine-actionable way. This means that (meta)data should be in databases that can expose it in such a way to allow search engines and harvesting servers to discover it, index it and link it to other relevant contextual information, thus vastly enhancing the likelihood of reusing the data (see Horizon Europe DMP template).


  • During a research project, scientists and researchers should utilize metadata in order to use, reuse, and expand knowledge by:
    • Managing experiments, samples and analysis pipelines.
    • Expanding current datasets e.g. to increase the sample size.
    • Repeating experiments done by colleagues in the team.
    • Reproducing and confirming findings done by others.
    • Testing new hypotheses on data generated for different purposes.
  • Scalable reuse of existing data is possible only if (meta)data is annotated with commonly used terms and findable by computers (e.g. database browser or search engines). The alternative could be very tedious and inefficient since you might have to:
    • Read the lab book of previous colleagues until you find the right page(s) where information about previously generated data is provided.
    • Look through numerous (shared) folders to find the documentation about a specific experiment done by previous colleagues that generated the dataset you are interested in.
    • Read all publications about a topic and check if there is a dataset linked to it and/or available upon request.
  • Integration of multiple datasets can be straightforward only if each dataset can be easily queried, processed and formatted via software/programmes that can properly handle structured and big (meta)data files, such as OpenRefine and programming languages such as Python or R. Otherwise, manual data integration and processing can be very slow and error-prone.


The advantages of having machine-actionable data and metadata are numerous for all the parties involved.

For researchers

By providing structured metadata and data to a database that follows standards (metadata schemas, ontologies, file formats, programmatic access, etc.), at the level of each recorded value or observation, researchers:

  • Could more easily query and filter (meta)data based on specific variables, experimental conditions, biological sources and many other parameters, based on the capabilities of the used ELN or data management software.
  • Can more easily find and reproduce experiments performed in the past by others in literature or in databases e.g. by using Europe PMC and EBI Search.
  • Can easily integrate data from multiple datasets and studies, sharing the same experimental conditions or variables. Datasets integration and manipulation are easier to achieve, more reproducible and can be automated by using common programmes/software such as R and OpenRefine.
  • Can make use of visualization and exploration tools, provided by some repositories, to browse and explore the data of multiple datasets at once. For instance, you can use Expression Atlas to easily make a query about the expression of a gene in specific conditions, even without knowledge of any data analysis software. As another example, GISAID allows you to visualise the spreading of viral variants. See the pages in the “Your Domain” section to find domain-specific databases, atlas or portals.
  • Can import, export and exchange (meta)data between tools/systems/platforms without data loss. Exchanging and integrating (meta)data between two software or platforms is possible only if the format in which the information is contained can be read and interpreted by both. For instance, (meta)data from both UniProt and PDBe-KB can be accessed in 3DBioNotes to enrich the structural analysis with sequence features.
  • Can explore and visualise biological knowledge graphs by using software such as KnetMiner and AgroLD.
  • Can perform complex queries, from a single entry point, across multiple distributed databases and across domains via APIs or via SPARQL Query Language. For instance: “Retrieve the number of UniProtKB/Swiss-Prot human enzymes that metabolize cholesterol or cholesterol derivatives and that are involved in diseases?” in the Integrated Database of Small Molecules.
  • Can more easily find reference data and existing data in general, since machine-actionable (meta)data could be found by search engines and domain specific or generic data catalogs and portals.

For software developers and repositories managers

  • Implementation of domain specific metadata schemas and ontologies for data and metadata increases the reusability for researchers.
  • The use of machine-actionable formats and ontologies contribute to syntactic and semantic interoperability of the (meta)data, which can be used by other tools/software or platforms.
  • Applying RDF syntax to the database can make the (meta)data available for knowledge graphs and semantic web applications.
  • If Application Programming Interface (API) is available, other software/applications could make complex queries, access the database programmatically and always get up-to-date data.
  • If the metadata of your database or repository is exposed according to specific standards, it could function as data provider or data source, and be harvested and indexed by
  • Machine actionable metadata facilitates the automatization of data handling and validation, allowing for easier development of new tools and analysis strategies (e.g. data visualization tools, machine learning and artificial intelligence applications).

For the authors of a machine-actionable public dataset

  • High impact of the published data.
  • More citations.
  • More opportunity for collaborations.
  • Opportunity for reproducibility test and confirmation of their results by others.
  • Easy way to reuse the same (meta)data for other research.
  • Improved scalability of their research efforts.

For public funders and institutions/governments

  • Proof that their fundings produced knowledge that is findable and reusable.
  • Transparency.
  • Straightforward collection and indexing of research output in registries for easier impact assessment and report.

What makes a file machine-actionable?


Due to the complexity of the topic and the lack of a unified definition, it is often difficult to identify the characteristics that make information contained in a digital object machine-actionable. Moreover, it is important not only to make a digital file machine-actionable, but also interoperable between different machines, so that different systems can exchange information.

The theoretically most machine-actionable format is in practice not achieved or established yet, however we try to list here some of the currently accepted best-practices that should be considered when making a file machine-actionable and interoperable, in Life Sciences research.


For machine-actionability and interoperability, you should consider:

  1. File formats that are data exchange formats (e.g. JSON, XML).
  2. (Meta)Data schemas recognised and accepted by communities as standards (e.g. ISA model, OME data model). The (meta)data schema describes the relations, such as hierarchy, of the elements that constitute the (meta)data model or structure.
  3. Sets of metadata attributes or metadata checklists recognised and accepted by communities (e.g. MIAPPE, ENA Samples checklists), that capture reporting best practice in the field.
  4. Controlled vocabularies and ontologies recognised and accepted by communities to convey meaning or semantics (e.g. EFO, OBI).

File format

  • Information contained in a digital object is only as accessible as the file format it is saved in. A file format is the way that information is encoded for storage in a computer file. This is often indicated by the file extension, e.g. .csv for a CSV file. Different file formats are preferred for different purposes. For instance:
    • PDF is tailored to store and display text and graphics to humans, according to specific layouts, but it is not suitable for exchanging information between machines.
    • CSV is appropriate to exchange plain text information in a tabular format, but its flat nature makes a challenge to describe more complex relationships between information.
    • XML and JSON formats are widely used for data exchange between systems (such as softwares, platforms or hardwares) and on the web. Both are easy to read and interpreted by machines, but not very human-readable.
  • Exchange file formats using key-value pairs, such as .xml and .json, can wrap or encode the information to be sent or exchanged in a hierarchical (tree) data model.
    • XML is a markup language. It is based on “elements” enclosed by pairs of “tags” and “attributes” (<tag>element<tag>). It is self-explanatory because it contains metadata about the format and “tags” are chosen by the creator of the .xml file. For instance, <name>Jaguar<name>.
    • JSON format can be easily read in any programming language. It is based on key-value pairs separated by colons ( {key:value} ). For instance, { name: “Jaguar” }.
  • File formats (or file extensions) for expressing data in triplets (e.g. “Jaguar” → “is in” → “Jungle”) in the Resource Description Framework (RDF) data model are .rdf for RDF/XML, .jsonld for JSON-LD, .nt for N-Triples and .ttl for Turtle syntax.

(Meta)Data schema

  • A (meta)data schema describes the relations, such as hierarchy, among the elements or pieces of information that constitute the (meta)data model or structure.
  • The relationship between pieces of information in a (meta)data schema can be implicit, following an agreed order (such as the order of columns in a table), or explicitly expressed by additional information in the file. To allow more universal interpretability, explicit additional information on the relationship between the pieces of information is highly advantageous.
  • Some of the (meta)data schemas considered standard in Life Sciences define the relations between elements of the model in a more implicit way (e.g. ISA-TAB, MAGE-TAB).
  • Some data repositories develop customised (meta)data schemas.
  • Different metadata schemas are preferred for different purposes. For instance,
  • The W3C consortium has formalised a universal abstract data model to potentially establish relationships among any resource available on the web (people, places, web pages, events, abstract concepts, etc) called Resource Description Framework (RDF). This universal abstract data model allows us to describe relationships between multiple resources encoded in different formats, following different standards and stored in different locations/servers on the internet.

    RDF model consists of sentences in the form of “Subject” → “Predicate” → “Object”, called Triples, that describe the relationship between different pieces of information. An example could be “Jaguar” → “is in” → “Jungle”. Subject and Object can be any resource available on the internet, Predicate (properties) connects resources to other resources or data values etc.

  • RDF concept can be written and applied to databases using different syntaxes, such as N-Triples, Turtle, RDF/XML, RDFa, JSON-LD. The benefit is that web browsers can put the provided information with these syntaxes into context and “understand” the meaning (semantics) and relations contained in the digital object. Information provided in RDF syntaxes is machine-interpretable. Digital objects in these formats can specify the context and the globally unique definition of each resource by referencing other standard metadata schemas and vocabularies/ontologies to describe web resources, such as or (for Life Sciences), Data Catalog Vocabulary (DCAT), Dublin Core, etc.

    Any metadata schemas and vocabularies/ontologies describing web resources can be expressed according to standards, such as the Web Ontology Language (OWL), the RDF Schema (RDFS) or the Simple Knowledge Organisation System (SKOS) to provide more expressive definition and inferences/relationships between terms or pieces of information.

Metadata checklist

  • Here, we define metadata checklists as content in the form of a fixed set of attributes or fields, without any particular order nor structure. Compliance to metadata checklists is not related to the format nor the structure, but rather to the content provided.
  • Many metadata checklists have been adopted as standards by Life Sciences communities (e.g. MIAPPE).
  • Some data repositories have customised metadata checklists (e.g. ENA Samples checklists).
  • Attributes in a metadata checklist can be ontology terms.
  • For more information see the Data documentation and metadata page.

Vocabulary or ontology

Vocabularies and ontologies are meant for describing concepts and relationships within a knowledge domain. For more information see the Data documentation and metadata page.


  • (Meta)Data in data exchange formats (XML, JSON, CSV, etc) that follows a standard metadata schema can be considered machine-actionable and syntactically interoperable. Ontologies that uniquely identify terms can be included for semantic interoperability.
  • RDF syntaxes, such as RDF/XML and JSON-LD, support syntactic and semantic interoperability among machines. In other words, these formats convey the structure of the data being presented and the link to the necessary information to interpret its content, e.g. ontologies. Ontology or vocabulary is a way of expressing semantics/meaning of (meta)data.

    Example of machine-interpretable metadata for the word “Jaguar” in JSON-LD format, which allows to clarify the intended meaning of the word “Jaguar” (the animal) and distinguishes it from other possible meanings such as car or computer:

    "@context": "", → "hey browser, I am using these definitions"
    "@type": "Taxon",
    "@id": "" , 
    "taxonRank": "species",
    "name": "Panthera onca",
    "alternateName": "Jaguar"
  • Metadata schemas and checklists can be found within RDA curated list of Life Sciences metadata standards or among standards in
  • Examples of standard (meta)data schemas, in different formats, in Life Sciences:
    • ISA-JSON (.json) and ISA-TAB (.txt) - generic metadata framework originally created to describe information about multi-omics experiments.
    • MAGE-TAB (.txt) - MicroArray Gene Expression Tabular. The format has been developed and adopted by the functional genomics community.
    • OME data model (.tiff or .xml) for a wide range of biological imaging modalities. Ontologies to uniquely identify terms can be included.
  • For more information about metadata schemas and ontologies, see Documentation and Metadata page.

More information

Relevant tools and resources

Skip tool table
Tool or resource Description Related pages Registry Bioschemas aims to improve the Findability on the Web of life sciences resources such as datasets, software, and training materials Data Steward: infrastructure Training
CEDAR CEDAR is making data submission smarter and faster, so that scientific researchers and analysts can create and use better metadata. Documentation and metadata Researcher Data Steward: research Tool info Standards/Databases
COPO Portal for scientists to broker more easily rich metadata alongside data to public repos. Documentation and metadata Researcher Plant sciences Tool info Standards/Databases
Data Catalog Vocabulary (DCAT) DCAT is an RDF vocabulary designed to facilitate interoperability between data catalogs published on the Web. Data Steward: infrastructure
FAIRDOM-SEEK Data, model and SOPs management for projects, from preliminary data to publication, support for running SBML models etc. Data storage Data Steward: infrastructure NeLS Microbial biotechnology IFB Tool info Training
FAIRDOMHub Data, model and SOPs management for projects, from preliminary data to publication, support for running SBML models etc. (public SEEK instance) Data storage Researcher NeLS Documentation and metadata Microbial biotechnology Standards/Databases
ISA-tools Open source framework and tools helping to manage a diverse set of life science, environmental and biomedical experiments using the Investigation Study Assay (ISA) standard Data Steward: infrastructure Data Steward: research Microbial biotechnology Standards/Databases
ISA4J Open source software library that can be used to generate a ISA-TAB export from in-house data sets. These comprises e.g. local database or local file system based experimental. Plant sciences Tool info
Research Object Crate (RO-Crate) RO-Crate is a lightweight approach to packaging research data with their metadata, using An RO-Crate is a structured archive of all the items that contributed to the research outcome, including their identifiers, provenance, relations and annotations. Documentation and metadata Data storage Data organisation Data Steward: research Researcher Microbial biotechnology Standards/Databases
Rightfield RightField is an open-source tool for adding ontology term selection to Excel spreadsheets Researcher Documentation and metadata Data Steward: research Microbial biotechnology Identifiers Tool info is a collaborative, community activity with a mission to create, maintain, and promote schemas for structured data on the Internet, on web pages, in email messages, and beyond. Data Steward: infrastructure Standards/Databases Training