Hello, community! As my first post in the OHDSI forum, I would like to introduce a research project I’ve been working on related to OHDSI vocabularies, and any help or feedback would be greatly appreciated!
I work for the University of Leicester; however, this research is my own. I’ve been building a terminology server that offers multiple vocabularies, ontological exploration, semantic similarity search, auto-completion, and related workflow components using an efficient API. It currently supports multiple vocabularies, including ontologies like HPO and SNOMED, genetic symbols like Ensembl and HGNC, and pathways like Reactome. As a recent addition, it now supports OHDSI.
However, unlike many other ontologies, OHDSI vocabulary is actually many ontologies combined into one, where each concept and relationship receives an OHDSI ID but still keeps its own original properties. I wonder if there is an official mapping between the different concepts, relationships, etc. of various sub-vocabularies inside the OHDSI vocabulary? Is my current implementation correct or efficient, or does it have any important issues?
I’m not sure whether this is directly relevant, but I’ve been looking at a related issue in terminology mappings: cases where a mapping appears reasonable but does not fully preserve the original meaning or granularity.
A recent paper proposes a new measure called the “Mapping Burden Ratio” to quantify differences in granularity between terminology systems.
I thought it might be relevant here, since semantic similarity does not always imply mapping equivalence.
OHDSI uses Athena as its primary resource for distributing vocabularies. OHDSI Standardized Vocabularies are also available elsewhere, but Athena is the central repository where we deliver the latest versions.
Have you been in contact with the Vocabulary Team already? We have a dedicated working group to discuss all things related to vocabularies. In August, the focus is on the upcoming release, but you are (obviously) welcome to join the discussion. The link to the working group is available here.
Hello @sohyeon , thank you for your response! I will definitely look into this, since I’m designing a mechanism to measure the mapping by embedding them in the same latent space. I’m encountering various difficulties along the way, though. This approach looks like a better fit for the question!
Hello @zhuk , thank you for your response! I haven’t been in contact with the vocabulary team; in fact, I didn’t know about the contact until just now. I would definitely be interested in talking with them to find out more! I’m attending the workshop happening soon in Birmingham, UK, and hopefully I can gain some clarity there.
As for the service, unfortunately, that was a courtesy hosting of the university, so it’s not very stable… I’ve fixed it and reported it to the hosting team; hopefully it will be better from now on. Thank you for reminding!