Closed sstemann closed 8 months ago
from TAQA: aragorn/robokop does do this BTE does Unsecret does.
in architecture - we will merge the "requirement/rule" in and then close this issue.
while i this the issue to enforce the requirement is merged, when do we expect to be able to verify which ARAs have met this requirement in Test?
And to chime in on this: while BTE and ARAGORN report they are returning look ups in addition to creative mode queries, in practice, they are not. O&O work shows only about half of all disease queries result in one or more drugs returned having at least two ARA's supporting it (and just 4% of all drug-disease pairs having multiple ARAs supporting it). It may be that the known treats are "falling off the edge" of the results when ARAs truncate results.
It may be that the known treats are "falling off the edge" of the results when ARAs truncate results.
I'd have to look at specific examples to be sure, but this is definitely known behavior for BTE. We definitely do the lookup, but there is no guarantee that all the lookup results are included in our top 500. I don't think it's always a bug per se, but it is something that will become less frequent when we complete our solution to https://github.com/biothings/biothings_explorer/issues/634.
@capasfield - is this something that can find its way onto the architecture agenda? :) I think it can be closed, but we should probably check with folks directly first?
Aragorn -Yes BTE - Yes ARAX - Working on it, Penn State is working on the issue Unsecret - Yes Improve - Challenge in ranking, but returning results
We definitely do the lookup, but there is no guarantee that all the lookup results are included in our top 500.
This raises again the interesting question of how Results that are based on a lookup / assertion / known fact should be scored/ranked relative to Results that are not (i.e. are only supported by creative prediction). If our scores/ordering is meant to reflect confidence that the results is correct, I would argue that results based on looked up known facts should (almost) always score higher than those based only on creative inference/prediction.
If/how/where we factor into a Results final score the 'knowledge level' of edges on which it is based - is an important but overlooked question that has come up in several settings.
look ups are included with inferred results. it may not be all ARAs are including both but marking this done, can open specific ARA issues as needed.
Per TAQA at the RENCI June Relay, and Architecture https://github.com/NCATSTranslator/TranslatorArchitecture/pull/88