QA of GO-CAM ingest

[mbrush]

QA Resources:

• KGX Summary report: https://docs.google.com/spreadsheets/d/1f6i3ASkjdrT6uEOkis82vaJofqJdrbLyzWiyhA43a-k/edit?gid=943440447#gid=943440447
• RIG (original): https://github.com/NCATSTranslator/translator-ingests/blob/main/src/translator_ingest/ingests/go_cam/go_cam_rig.yaml
• RIG (PR with QA changes)

---

Issues to consider / address for next release:

Generally, the initial release of this was pretty fast and loose, and there are plans to come back and improve things. Lots of notes to this effect in the RIG 'future considerations'. Noting some additional observations below:

• uses predicates that violate Translator conventions (e.g. include direction of effect)
• very few edges actually created
• those that are lack detail e.g. qualifiers, edge properties that could be taken)
• contains relatively large numbers of nonsensical edges (e.g. gene causes gene (878), gene has input gene (1400)

Also, the RIG is not in sync with what the code is actually doing to produce data, and seems to reference predicates that are not in Biolink (e.g. positively_regulates), and edge properties that are not in biolink (apparently as placeholders to capture qualifying info that will be captured using proper SPOQ patterns in future iterations of the ingest)

---

Issues to consider longer term:

• Some Biolink predicates include directionality (e.g. acts upstream of negative effect) - which we capture in qualifiers for Translator use cases. We should not use these in our ingest, and define qualifier based patterns to represent directionality of effects.

---

Questions:

[sierra-moxon]

"uses predicates that violate Translator conventions (e.g. include direction of effect)" - I do not think this is true, except for "acts upstream of..." predicates which are longer term modeling decisions we need to make. these are quite consistently used in GO.

"very few edges actually created" -- these are GO-CAMs. not go-annotations. there are not going to be a lot of edges, there are not a ton of GO-CAMs that we release for this purpose.

"those that are lack detail e.g. qualifiers, edge properties that could be taken)" -- true.

[RichardBruskiewich]

@sierra-moxon and I are cooperating on an open Biolink Model issue: biolink/biolink-model#1719 reviewing possible Biolink Model revisions to enhance GO-CAM ingest semantics.