FAQ: Why LinkML
Why should I use LinkML at all?
All data follows some kind of schema or data model, whether it is explicitly articulated, or implicit / in the background. In our experience it is always beneficial to explicitly articulate that schema. This holds true for a range of situations, including:
you have simple tabular data in TSVs or a spreadsheet
you have a highly interlinked relational model
you are working with JSON documents in a document store like MongoDB
you have a knowledge graph in Neo4J
you are working with linked data in RDF
LinkML is designed to be flexible enough to cover all these use cases, allowing for lighweight semantic data dictionaries for tabular data, through rich interlinked schemas for knowledge graphs and triplestores
My data is a simple spreadsheet/TSV, why should I use LinkML?
If your data is a simple CSV then using a framework like LinkML may seem daunting. It may be tempting to include a simple README alongside your TSV describing the column headers. While this is certainly better than not having any data description at all, consider some of the drawbacks:
the README isn’t a computer-digestible representation of your data
you do not have a mechanism for validating the TSV
you lack a computable way of mapping your column headers to a standard data dictionary, vocabulary
LinkML provides an easy way of making your data computable. You can start with a very simple schema, with a single class, and one slot per column in your TSV; for example, for a dataset where the rows represent people:
classes:
Person:
attributes:
id:
name:
email:
age:
occupation:
...
this is already useful as it states which column headers are expected in your data.
you can later expand that to provide constraints on the values each column:
classes:
Person:
attributes:
id:
identifier: true ## this ensures the id field is unique
name:
email:
pattern: "\\S+@[\\S\\.]+\\S+" ## regular expression
age:
range: integer
minimum_value: 0
maximum_value: 999
occupation_class:
range: job_code ## enumeration
unique_keys:
- description: email is unique
unique_key_slots:
- email
enums:
job_code:
scientific:
technical:
service:
You can also provide textual descriptions and additional metadata on the columns that may be useful for human users
classes:
Person:
attributes:
id:
description: unique identifier
name:
description: the full name of the person
email:
description: the persons email address
age:
description: the age of the person in years
occupation_class:
description: the kind of job the person has
...
If you like, you can provide both linked data IRIs to your schema elements, or mappings to existing systems, for example:
prefixes:
schema: http://schema.org/
foaf: http://xmlns.com/foaf/0.1/
classes:
Person:
attributes:
id:
slot_uri: schema:identifier
name:
slot_uri: schema:name
email:
slot_uri: schema:email
exact_mappings:
- foaf:email
This has a number of advantages:
you make the intended meaning of your columns more transparent and explicit
your TSVs can be automatically translated to JSON-LD and RDF via the LinkML framework
it faciliates automated and semi-automated mapping between your data and other representations
There are a number of proposed frameworks for providing lightweight data dictionaries for your TSVs/spreadsheets:
frictionless table schemas
csvy
csv on the web
One advantage of LinkML is that it is not only for TSVs
Why should I use LinkML over JSON-Schema?
JSON-Schema is a fantastic framework for validating JSON documents. If your primary use case is validating JSON documents then by all means keep using it!
However, if any of the following apply to you, you may want to consider LinkML - and remember, you can always compile your LinkML schema down to JSON-Schema!
You want to make use inheritance/polymorphism
you want a language with a simple core based around familiar concepts of classes and fields (slots)
you want to make your data more FAIR (Findable Accessible Interoperable Reusable), for example, by annotating schema elements with IRIs
you want to use JSON-LD but don’t want to coordinate separate JSON-Schema and JSON-LD context documents - LinkML is an all-in-one-solution!
you want a more advanced ontology-aware enumerations
you want your datamodel to be used in other frameworks than JSON - e.g. TSVs, SQL databases, Triplestores, graph databases
When making your decision, you should weigh factors such as the fact that things that can be expressed in one framework may not be expressible in the other.
See also FAQ entries in modeling which compare some similar constructs.
Why should I use LinkML over ShEx/SHACL?
ShEx and SHACL are both excellent “shape” languages for providing closed-world constraints on top of RDF/triples. If your universe consists entirely of RDF then you may want to consider adopting one of these as your primary means of expressing your data!
However, if any of the following apply to you, you may want to consider LinkML - and remember, you can always compile your LinkML schema to ShEx, with the ability to do this for SHACL coming soon!
you want your datamodel to work for non-RDF representations such as JSON, YAML, and relational databases
your user based and developer base is not entirely semantic web / linked data enthusiasts
your emphasis is more on data modeling rather than validation
Full disclosure: LinkML lead developer Harold Solbrig is also one of the authors of the ShEx specification.
Why should I use LinkML over SQL DDL?
SQL Data Description Language (DDL; or simply “CREATE TABLE” statements) is a means of describing the structure of a relational database.
If you are using a relational database and have minimal lightweight application code that performs direct SQL queries over data, there may be no compelling use case for LinkML.
However, most information architectures that use a SQL database also involve some alternative representation of the data - for example, a JSON representation in an API, or an object representation – or even an RDF representation. It can be challenging to keep these different representations in sync. There are frequently excellent products that solve “one part” of this mapping problem – e.g. an ORM (Object Relational Mapping) tool. LinkML is intended to allow you to give a “bigger picture” view of your model that is as independent as possible from underlying storage or exchange technologies, and at the same time can be compiled down.
LinkML also allows you to specify URIs and CURIE/URI mappings for each element of your schema, which provides a declarative specification of how your data should be mapped. For example, if you are trying to bring together two schemas (whether via federation or via warehousing), if both schemas annotate the same field with the same URI then we know these can be merged.
Even if you are not interested in mapping your SQL data model to anything else, it can still be a great idea to use LinkML as a data definition language for your schema, especially if you have a schema with many fields that a domain scientist or stakeholder needs to understand.
In our opinion, SQL DDL is not a great language for data dictionaries. There is a lack of standard ways to even add comments or descriptions to fields, and it can be challenging to introspect these. LinkML provides a simple easy to use way to provide rich metadata about your fields.
Compare:
CREATE TABLE sample (
id TEXT PRIMARY KEY, -- unique sample id
individual_id TEXT FOREIGN KEY(person.id),
name TEXT,
disease TEXT,
src TEXT,
collec_location TEXT FOREIGN KEY(geoloc.id),
...
);
with:
classes:
Sample:
attributes:
id:
title: identifier
identifier: true
pattern: "SAMPLEDB:SAM\\d{8}"
description: A unique identifier for the sample
name:
title: sample name
description: A human-readable name for the sample
range: NarrativeText
disease:
title: disease
description: the disease with which the patient was diagnosed
range: DiseaseEnum
src:
title: tissue source
description: the anatomical location from which the tissue was sampled
range: DiseaseEnum
todos:
- model this using an ontology term instead
collec_location:
title: collection location
description: the geocoordinates of the site where the sampling was obtained
notes:
- this should NOT be the site of processing
range: GeoLoc
Here we have a standard way of providing human-readable names for our columns (via title). We have a human-friendly textual description. Both of these could be used to drive dynamic tooltips in an application that collects or displays data.
Why should I use LinkML over UML?
TODO
Why should I use LinkML over OWL?
LinkML is in a very different class of languages from OWL. LinkML is a schema language, with features in common with JSON-Schema, XML Schema, UML, SQL DDL, shape languages such as ShEx/SHACL.
In our experience OWL is best for “open model” direct representations of domain knowledge, and should not be used as a schema language. However, we appreciate this is nuanced, and we welcome any questions or discussions via our GitHub issue tracker.
If you do have a schema expressed in OWL, you can use the linkml toolkit to infer/bootstrap a LinkML schema from it. But note this is a heuristic procedure, and will not give you sensible results from a large “terminological” ontology.
It is possible to use LinkML to help you structure an OWL ontology by using LinkML as a metaclass authoring system. See ChemSchema for an example, and see also the linkml-owl framework.