search

pantherdb / fullgo_paint_update

Update of Panther and PAINT DBs with monthly GO release data
0 stars 0 forks source link

Prevent IBA propagation to leaves having conflicting qualifiers (NOT) #30

Open dustine32 opened 5 years ago

dustine32 commented 5 years ago

During IBA GAF creation, check GO annotation qualifiers on leaves when deciding whether to create IBA. If the IBD GO term matches with experimental leaf GO annotation term but qualifiers don't match, do not create IBA to that leaf.

Will find some examples to test.

dustine32 commented 5 years ago

From https://github.com/geneontology/go-annotation/issues/2660: image IBAs for SPAC1B3.15c to GO:0015225 do not have NOT qualifiers, which conflicts with the experimental annotation. This is likely due to the input files (go_aggregate) for createGAF.pl containing qualifier data that ALSO conflicts with what's in our DB. This then is loaded into the perl script's data structures incorrectly so that it looks like there is no qualifier conflict despite the output IBA qualifiers differing.

The cleanest (but likely more labor-intensive) approach is to change the monthly PAINT GO update pipeline to ensure imported GO annotation qualifiers are correctly paired with each annotation's evidence code. Will do some pipeline-load testing to get this right.

dustine32 commented 5 years ago

To clarify, this ticket was originally intended to only prevent IBA propagation if the qualifiers for an experimental GO annotation and an IBD using it as evidence do not match, specifically, "has NOT qualifier" vs "has no qualifier." If these qualifiers don't match and and there is no other experimental evidence that can support the IBD, the IBD will be obsoleted thus preventing IBA propagation.

I believe, other than the redundant IBD bug in issue #35 , this is fixed in the PAINT update pipeline.

Another case for blocking IBA propagation is when an IBA is attempting to be propagated to a leaf sequence that has an experimental GO annotation with conflicting qualifiers (again, only checking the "has NOT qualifier" vs "has no qualifier" case). In this case the IBA should not be made for this leaf. The IBD can still be valid and the blocking effect would just apply to that leaf. The above image in comment is a good example. This has yet to be implemented in the IBA generation script but I'll make a new ticket and get it going for the end-of-November PAINT update.