Set Interpretation: Revisiting collation behavior

[tokebe]

(For historical context see the previous collation issue)

Current v1.6 set interpretation leaves behind a special behavior in the old is_set schema. Previously, one could set is_set: true on a query node which presents no IDs in order to signal that results should not be uniquely identified by that node; where each result should collate all applicable IDs matching that given query node.

(For instance, Drug -interacts_with-> Gene (is_set: true) -causes-> Diabetes would lead to a result Drug A -interacts_with-> Gene [A, B, C] -causes-> Diabetes)

This is not currently expressible in TRAPI v1.6, where set interpretation MANY and ALL both require that member_ids is set to an enumeration of IDs.

The comments below will list some possible solutions, which we can discuss and vote on later.

[tokebe]

Option 1

Add a new set_interpretation value: COLLATE, which may only be set on a node which provides no IDs, and signifies that the above-described behavior is desired.

[tokebe]

Option 2

Amend the descriptions on QNode such that set_interpretation: MANY may validly be used on a node which provides no IDs, and signifies the above-described behavior is desired.

[tokebe]

Option 3

Add a top-level field named collate on the QNode that signifies the above-described behavior is desired. This opens a question about how collate should interact with set_interpretation, most likely the answer is that only one may be used.

[colleenXu]

NOTE!

I think we may need to adjust some more QNode property descriptions because they all mention set_interpretation behavior:

• ids
• member_ids

[colleenXu]

Documenting 2/12 outcomes: discussed related standards (Engineering 12.iii), reached consensus on behavior and wording.

Option 1 implemented in PR #543 and merged into 2.0 branch. So closing issue.