2021-12-08 In person meeting at the Squealing Pig

[explodecomputer]

Present: Gib, Rafael, Vince, Robert

Prior to the meeting Tom generated parsed trait names and sent to Rafael, and Rafael used the ontology mapping tool to generate a list of traits that do not map here: https://docs.google.com/spreadsheets/d/1BUmhJO9f6woSuMnF5YfEwgLWR9p-xCQXMzh_s-S3yK4/edit#gid=639951542. Gib went through each row to develop a strategy going forwards.

Meeting notes:

We develop two ontology mappings for OpenGWAS. a) the narrow mapping: the ontology matches strongly to the intended meaning of the trait definition. b) broad mapping: where the trait is too abstract or uninterpretable we map it to an ontology as a broad category. Traits will only have narrow or broad mappings. If a trait needs a narrow mapping but it is not present in EFO then we request it. Narrow mappings can be mapped to broad categories automatically through the ontological hierarchies.

Next steps:

1. Develop a document that explains how to do ontology mapping
2. Recruit volunteers / mappers

Other potential use cases for ontologies

1. Search function by navigating along the ontological hierarchy
2. Use the ontology as a prior for improving power of genetic association at child ontologies that share a parent ontology.
3. Use ontologies in automated multivariable MR - we want to identify the set of exposures that share instruments and automatically analyse them in MV-MR. However, that might accidentally include proxies for the outcome. Use ontologies to remove any proxies.
4. Traits can be clustered by navigating along the ontological hierarcy. It would be interesting to see how ontological clusterings correspond to the clusterings from embeddings e.g. in vectology.