The term knowledge graph was first introduced by Google in 2012. If you have ever completed a search using the engine (which you almost certainly have!) then you have consumed data served by a knowledge graph; it’s the underlying graph structure that populates the box on the right hand side of the results page. Google’s knowledge graph harmonizes data from a number of public sources to provide a comprehensive summary of the query entity. Other large technology companies also make use of this data representation, including Facebook’s social network and the Amazon product graph. This is all very well, but it still doesn’t provide us with an answer to what exactly is a knowledge graph?
In an oh-so common scenario within the field of technology, there exists a plethora of definitions to describe a knowledge graph. These definitions not only range in clarity and complexity but are used interchangeably and are often only meaningful to the area of application. A safe and simple definition of a knowledge graph that we use is…
a semantic graph that integrates information into an ontology
In a graph representation, entities or ‘things’ are represented as nodes, or vertices, with associations between these nodes captured as edges, or relationships. Furthermore, nodes and edges may hold attributes that describe their characteristics (see Fig 1.).
The fact that a knowledge graph is semantically enriched means that there is meaning associated to the entities in the graph, i.e. they are aligned to ontologies. For example, a node that has the name NASH is pretty meaningless in and of itself. To a scientifically knowledgeable human it may be clear that this node refers to a disease, but how would a computer assign a type to this node; is it a gene, a drug or even a person? Furthermore, which other nodes this may interact with and via what type of edge? A knowledge graph gets around this by labelling the NASH node as a disease; by aligning this node to a disease ontology a computer can start to understand that entity in the context of other node types that may also be in the knowledge graph. Simply put, a knowledge graph understands real-world entities and their relationships to one another: things, not strings.
If we also have genes in the graph we can add edges between diseases and genes that describe associations in the form GENE -> associated with -> DISEASE (see Fig 1.). Read more in our use case on using phenotype triangulation to improve disease understanding.
Ok, so we now have a definition of a knowledge graph but what makes this data representation so powerful?
A knowledge graph can be used to connect data from numerous heterogeneous data silos, whether they are external or internal, provided entities are harmonized to common identifiers – something we will touch on shortly! Unlike more restrictive relational databases, graphs allow for the creation of typed relationships with attributes attached in a lot more intuitive a representation than foreign keys or join tables. Graphs don’t rely on prohibitive schema and can be updated and modified as and when required, as a project evolves.
Furthermore, when aligning data in your graph to ontologies, as well as the semantics, you also get the metadata captured in the ontology for free. Finally, once your data has been integrated into a single view, inferences that would have been otherwise unseen can be made. We have also seen in recent years that the technology supporting knowledge graphs has matured and, importantly, is scalable. Graph databases, with intuitive query abilities, have reduced the barrier for entry for those interested in knowledge graphs dramatically.
In order to get the most out of your knowledge graph it’s important to understand the use case you are trying to address from the offset. Typically speaking, there are 2 approaches to creating knowledge graphs, at an enterprise level for search, or at a project level to enable inferences.
An enterprise knowledge graph will, by definition, be more abstract, including data from many departments in a company, e.g. finance, HR, legal, R&D etc, where everybody is viewing the data from a different aspect, or through a particular lens.
At a project level, a use-case is more clearly defined- what specific questions do we want to ask of the knowledge graph? For example, an exercise in target prioritization will require focus on gene-disease associations and relevant datasets. In this particular use-case the addition of HR documents will likely not help!
We have described what a knowledge graph is, what makes a knowledge graph so powerful and the importance of identifying a use-case, but what can we at SciBite do to help facilitate the production of these I hear you ask! This facilitation can be broken down into 3 areas…
The above functionalities provide the basic ingredients required for a knowledge graph pipeline. By knitting the pieces together in a connected workflow, you can start to see how SciBite can support the creation of ontologies while also harmonizing and integrating data from both unstructured and structured data sources; aligning such data to the supporting ontologies. Such a pipeline could be semi-automated or even automated, depending on the use case.
The great news is SciBite’s knowledge graph facilitation with data harmonization/extraction is completely agnostic to the technology you wish to use to represent or indeed store your knowledge graph. So whether you are an RDF expert (check out our blog on SciBite & RDF – A natural semantic fit) looking at triplestores supporting SPARQL endpoints or more interested in the ease that comes with LPGs and the more intuitive graph query languages that come with these, SciBite can help you…
Watch our webinar on Creating Knowledge Graphs from Literature to learn more.
If you have any questions regarding the use or facilitation of knowledge graphs, or regarding a particular use case, please don’t hesitate to get in touch and one of our experts will get back to you.
In the first of this two-part blog, I describe what ontologies are and how you can use them to make the best use of scientific data within your organisation.Read
In this blog we discusses how Sherlock Holmes (amongst others) made an appearance when we looked to exploit the efforts of Wikipedia to identify articles relevant to the life science domain for a language model project.Read
Get in touch with us to find out how we can transform your data
© Copyright © 2023 Elsevier Ltd., its licensors, and contributors. All rights are reserved, including those for text and data mining, AI training, and similar technologies.