Closed dosumis closed 1 year ago
@dosumis, please describe an action item and acceptance criteria for this ticket. As it is, I'm unsure how to proceed.
Use your judgement - given the specified use-case.
Also see SOP ticket. https://github.com/obophenotype/cell-ontology/issues/1919
Use your judgement
Thanks for the link.
My pending question is: Use judgement to do what exactly? I would need some acceptance criteria to know what the revised intended goal is.
The originally submitted list of classes (before the removal of overlaps) was my determination of a balance between overlap and specificity. If you can specify metrics that would constitute an improvement on that list, I can re-review the original list against those clarified metrics.
Dropping all classes with score of 1 and replacing with the parent class seems like a large gap in granularity in some cases, e.g. platelet and 'myeloid cell'. Can you provide a metric on how to determine which of those two terms would be appropriate?
Happy to discuss offline and record the clarified action item and acceptance criteria in this ticket.
before the removal of overlaps
I don't think overlaps have been removed.
before the removal of overlaps
I don't think overlaps have been removed.
@anitacaron, can you confirm the blood_and_immune_upper_slim overlap terms will be removed once #1939 is merged? Based on this comment, it seems like they will be, but can you confirm?
@bvarner-ebi, @ubyndr put them back on this commit and removed the QC for overlapping classes, requested by David offline.
All action items appear to be addressed. If anything else is required, kindly reopen with required action items.
The first round of generating the blood_and_immune_upper_slim slim was done without the help of reporting tools. Now those tools are in place, I think we can see some room for improvement, and some ways we could write a better SOP.
The main use case for an upper slim is to summarise data. Things that potentially get in the way of this use case:
Potential clashing concern: It seems reasonable to want to make sure that very important cell types are not obscured in generating summaries, but this desire can clash with the considerations above. Some judgement may be needed.
Report for current slim
Overlap report (generated by Anita
query: https://api.triplydb.com/s/_b0O6UP2A
SOme conclusions:
These do not group (a score of 1 = no subclasses)
platelet,1 multinucleated giant cell,1 nucleated thrombocyte,1 natural helper lymphocyte,1 B-lymphoblast,1 lymphoblast,2 blood lymphocyte,2
Mononuclear cell has >400 subclasses and is a source of much of the overlap. Using OLS with coverage (subclass) counts displayed shows some good possibilities for choosing more specific terms.
Judgement required. There are no perfect solutions - but some are better than others.