BLOG
Data itself often has little to say, it’s the way we use it that defines its value. Ontologies are key players in structuring data in a meaningful way. Despite their importance, there is often a lack of clear understanding of what ontologies are and how to use them effectively.
Ontologies add context and meaning to data by defining categories and associations. Even if you have an excellent AI algorithm, a backbone for your data is essential to clarify the signals and get the most value of it. To effectively create an ontology, you need to understand the reasoning behind it and follow some essential steps. In this article, we will familiarize you with these different steps to guide you out of your data maze.
1. Controlled vocabulary
The first step is about reducing variability and consolidating related information. By creating generalized categories such as ‘drug’ and ‘disease’ and listing the authorized values, your data already becomes more meaningful and standardized. Synonyms can be grouped together, and typos can be identified, leading to an unambiguous dataset.
2. Hierarchy
After creating categories, subbranches refine your data. This allows you to adopt different perspectives on the information. For example, as a neuroscientist, you are probably more interested in data on neurodegenerative diseases alone instead of all diseases together.
3. Thesaurus
While the previous steps establish definitions and descriptions of the datapoints, the third step identifies associative relationships between them. An antibody can have a link with another drug modality, such as a small molecule, or it can be related to a certain disease. The nature of these relationships is, however, not yet identified.
4. Ontology
Lastly, accompanying the established relationships with consensus definitions structures your data as a knowledge graph. A drug can, for example, either be indicated or contra-indicated for a certain disease. These associations allow the implementation of machine learning applications, further increasing the value of your data.
Although it’s tempting to jump in headfirst, building an ontology is a journey. Each step builds the foundation for the next one. First, you need to ‘get to know’ your data and identify key concepts, hierarchy and relationships, and acquire domain knowledge through conversations with experts. Since this whole process is time-consuming, it’s important to emphasize that the final goal is not building an ontology per se. The priority lies in increasing the value of your data with each step along the way and focusing on what works best for you at this moment. Not every application needs an ontology, and a controlled vocabulary is already a huge leap. It is better to view it as a continuous process of updating your backbone as new data comes in and re-evaluating your needs.
Different tools exist to get you started on your journey. For a controlled vocabulary, an Excel file or simple database will suffice. For more advanced applications, dedicated platforms to visualize your ontology are recommended. Some examples include Protégé, which is free and open access, and CENtree (Elsevier), TopBraid (TopQuadrant), SmartLogic (Semaphore).
A good start is half the work: taking the first step and adapting your course along the way will help you obtain as much value as possible from the data you worked so hard for.
