Failure of -obo-report.tsv induced by OBO serialisation quirks

[gouttegd]

This problem was encountered by Uberon on this PR; see the CI failure report here.

The main change in that PR is the addition of a new term that is disjoint from two imported GO terms:

[Term]
id: UBERON:0000001 ! gross anatomical part
name: gross anatomical part
disjoint_from: GO:0005623 ! cell
disjoint_from: GO:0110165 ! cellular anatomical entity

(Other tags in that frame removed for clarity, they don’t matter here. What matter are the two disjoint_from tags.)

At some point in the standard pipeline, the -edit file is processed to generate what we call $(SRCMERGED). Basically, imports statement are removed and components and pattern-derived axioms are merged. Importantly, $(SRCMERGED) is in the same format as the edit file, so if the ontology is maintained in the OBO format (as is the case for Uberon), then $(SRCMERGED) is also written in the OBO format. $(SRCMERGED) is then used as input to robot report to generate the reports/uberon-edit.obo-obo-report.tsv report, which is the step that failed in the above PR.

The problem is that, when writing $(SRCMERGED), the disjointness axioms that, in the -edit file, are part of the UBERON:0000001 frame, are now written as if they were part of the frames of the referenced GO terms. That is, the fragment above becomes:

[Term]
id: GO:0005623
disjoint_from: UBERON:0000001 ! gross anatomical part

[Term]
id: GO:0110165
disjoint_from: UBERON:0000001 ! gross anatomical part

[Term]
id: UBERON:0000001
name: gross anatomical part

(I have not looked at the logic used by the OWLAPI to decide in which frame a disjoint_from tag should go, but I suspect the decision is made by lexicographically sorting the IRIs involved in the axiom, and putting the tag in the frame corresponding to the first IRI. In this example, GO:0005623 would be sorted before UBERON:0000001, so the OWLAPI creates a frame for GO:0005623 instead of writing the disjoint_from tag in the frame for UBERON:0000001.)

Now, logically, this does not change anything. A DisjointWith: B is obviously the same as B DisjointWith: A. But the second form makes the missing_label test (part of the standard ROBOT report check) fail, because the GO:0005623 and GO:0110165 frames don’t have a label!

[gouttegd]

The easiest fix, I believe, would be to always write the $(SRCMERGED) file in OFN format (or any non-OBO format, really), independently of the format of the edit file.