Refining creative-mode templates for Sept 2023 release

[colleenXu]

Making a new issue to track the work related to writing creative-model query templates (involves testing them against diseases in individual queries, and deciding the order to run them in....).

Related to previous comments in #449 and #453.

[colleenXu]

TDLR: There's still work:

• Does the corrected notebook need to run on the latest DrugMechDB data?
• Review short metapaths from that notebook for use as templates (cells 48, 51?) Max of 3 qEdges
• Write / refine templates. Remember design so that individual results within them == unique drugs (using is_set parameter to do this).

From Slack discussions:

• Andrew wanted me to start with Carolina's templates, assuming that these were related to an older DrugMechDB analysis (output of cells 20-21 here). However, these two didn't actually match each other
• Turns out Carolina's templates were based on Kevin's bte-regression-testing here, which doesn't seem to be based on DrugMechDB analysis
• Carolina ran a corrected DrugMechDB analysis on a "DrugMechDB version downloaded in April". I stored it here
  • What Carolina said about the notebook: The results from this notebook are wrong because the sequence of meta-nodes are extracted incorrectly from the DMDB path file. For example, in a path that only has one biological process identifier, the extracted meta-path shows two biological process nodes.

[andrewsu]

I'm going to close https://github.com/biothings/BioThings_Explorer_TRAPI/issues/237, but just noting here that orthology-based templates could also be good for drug-disease templates for creative mode...

[tokebe]

@colleenXu @andrewsu Just taking a brief browse through the TAQA triage board, looks like this issue may come from us. I think we could avoid this with some changes to make the directionality of m4-Chem-Gene-DoP.json more specific (possibly through qualifiers/etc). Though, based on @andrewsu's comment in another issue, it appears we may not want to, if I understand?

[colleenXu]

Seems like @andrewsu would be the one to decide...I agree that adding a qualifier constraint to the template could work to make this more specific (object direction is decreased)

[tokebe]

Based on today's meeting it seems we want to make the template more specific, but duplicate the less-specific version as a template with lower execution priority.

[colleenXu]

Keep in mind: UI has a 3-hop limit and subclass edges from node-expansion will count as 1. So templates should be only 1 / 2 hops right now...

[colleenXu]

Also, can modify templates now that "qualifier hierarchy expansion" is now supported. See https://github.com/biothings/biothings_explorer/issues/598

[colleenXu]

As discussed in today's meeting, the scope of this issue (defined for Sept release) is:

• make the directionality of m4-Chem-Gene-DoP.json more specific
• simplify templates now that qualifier expansion is supported

[colleenXu]

Originally, this issue was addressed by https://github.com/biothings/bte_trapi_query_graph_handler/pull/167.

But after #699 and #701 were identified, I adjusted the templates. The two main asks of this issue are still addressed by the adjusted templates + templateGroups.

Details of the adjustments made can be seen in #699 and #701 and in the PR that addresses them https://github.com/biothings/bte_trapi_query_graph_handler/pull/169#issuecomment-1683004440. I include basic info on the latest testing in this PR.

[tokebe]

Tracking will continue in linked and future issues, closing this one as completed