Home
Knowledge graphs are an important data structure in biomedical informatics providing a means for representing complex relationships between heterogeneous data sources. Within the Semantic Web domain, there are long-standing epistemological debates regarding the ways in which knowledge graphs should be constructed. The two predominant knowledge models used when constructing knowledge graphs include the instance-based (i.e. adding new entities as an instance of an existing class or instance. See GO-CAM - Gene Ontology-Causal Activity Models) and subclass-based approaches (i.e. adding new entities as a subclass of an existing class. See KaBOB - The Knowledge Base of Biomedicine).
Problem: While there are many examples of knowledge graphs constructed using either approach (or a combination of both), there are no empirical evaluations or comparisons of these approaches making it difficult to understand how they differ and when one method should be used over the other. To date, there are no existing libraries that enable the construction of knowledge graphs under alternative knowledge models.
Solution: To address this limitation, we developed PheKnowlator (pkt-kg), a Python library designed to facilitate the construction of large-scale biomedical knowledge graphs under different knowledge models. In addition to providing users flexibility in selecting the underlying knowledge model, pkt-kg also allows users to specify how relationships between entities are added (i.e. using a single relation or relation and its logical inverse) and provides options for generating alternative knowledge graph "views" or "representations" via lossless transformation into property graphs (e.g. see OWL-NETS).
(Click Figure to Enlarge)
Translational Research Informatics Team
 |
 |
 |
 |
 |
|---|---|---|---|---|
|
Tiffany Callahan 🖥️ Project Lead |
Bill Baumgartner 🖥️ | Ignacio Tripodi 🖥️ | Adrianne L. Stefanski 🔬 | Jordan Wyrwa ⚕️ |
Research Mentors
The resulting knowledge graphs and molecular mechanism embeddings are free to download and included as part of each release. See the following page for details: https://github.com/callahantiff/PheKnowLator/wiki/Archived-Builds.
- Methods manuscript preprint can be found here: https://doi.org/10.48550/arXiv.2307.05727.
- We presented this work (poster) at the 15th Annual Rocky Mountain Bioinformatics Conference
- Ignacio Tripodi will present results on MechSpy, a novel application that uses PheKnowLator to perform toxicological mechanistic inference at the 2019 meeting of The American Society for Cellular and Computational Toxicology
- PheKnowLator is referenced in a review article on Knowledge-based Data Science in the biomedical domain
- PheKnowLator was mentioned in a recent blog post
- The computational performance of PheKnowLator will be presented at the 2020 annual International Conference on Intelligent Systems for Molecular Biology (Prerecorded Talk; Submission)
We'd love to hear from you! To get in touch, please create an issue or send us an email.