BUG: robot template with class expressions duplicates classes as datatypes

[daniwelter]

From the ROBOT Slack channel:

I'm having a weird issue with robot template since the last release. It's probably user error but since I upgraded to v1.9.6, any classes used in class expressions become duplicated as datatypes. This is an extract from my template file:

ID	Type	Label	Parent
ID	TYPE	A rdfs:label	SC %
AfPO:0000277	class	Western African	HANCESTRO:0010
AfPO:0000410	class	Adele	(RO:0000086 some HANCESTRO:0599)
AfPO:0000325	class	Afrikaner	(RO:0000086 some HANCESTRO:0599)

HANCESTRO:0010 turns out fine but HANCESTRO:0599 is now both a class and a datatype. I've tried various permutations, e.g. having a column header of

Ethicity Descriptor
SC RO:0000086 some %

and then just HANCESTRO:0599 in the cell, but the outcome is the same. I also double-checked some components that have been around for a while, and the problem occurs for these as well when it hasn't before (comparing with the current release version from April 2024).

As per @jamesaoverton's suggestion, I then replaced RO:0000086 with 'has quality' (and the same for any other object properties in my templates) and this fixed the issue.

Problem replicated in release v1.9.6, 1.9.5 and 1.9.4. Last successful version uncertain but may have been 1.9.2 or 1.9.3.

[matentzn]

I bet this PR is somehow related to this issue: #1104, but I am not sure. Timeline fits.

[ramonawalls]

I am having the same problem. See https://github.com/PlantPhenoOntology/ppo/blob/odk-conversion/src/ontology/components/PPO_plant_structures.owl. The template is at https://github.com/PlantPhenoOntology/ppo/blob/odk-conversion/src/templates/PPO_plant_structures.tsv

I am using the latest ODK.

I can't replace RO:0000086 with 'has quality' as @daniwelter was able to do, because of the problem described in #1053.

Feeling sad.

[jamesaoverton]

Hi @ramonawalls. You should not need to use CURIEs everywhere in your templates. ROBOT is much happier if you use labels. And you should not need to quote labels most of the time -- only when they occur inside a logical expression. OBI has many examples: https://github.com/obi-ontology/obi/tree/master/src/ontology/templates.

In order to parse templates (whether labels or CURIEs) ROBOT needs to know that you're using Object Properties or Data Properties or Annotation Properties or Classes, etc. That's why the --input foo.owl option is important.

[ramonawalls]

Thank you again! I'm noting issue #442 so the examples only have to be in one place.

@jamesaoverton, could you please provide links to the corresponding input files used in the OBI examples?

[jamesaoverton]

This command runs the various OBI templates: https://github.com/obi-ontology/obi/blob/master/Makefile#L92. It imports obi-edit.owl, which imports most of OBI, including previous template results. @matentzn can also provide many examples.

[ramonawalls]

@matentzn here is an example of a template that generates duplicate datatypes and the corresponding OWL file:

https://github.com/PlantPhenoOntology/ppo/blob/odk-conversion/src/templates/PPO_plant_structures.tsv

https://github.com/PlantPhenoOntology/ppo/blob/odk-conversion/src/ontology/components/PPO_plant_structures.owl

[ramonawalls]

A couple more points.

1. I did try to use my edit file as input, but I couldn't get it to work after a few tries. It would be really nice to have an example (maybe in the ODK documentation) on how to specify a robot input file in the yaml template for the ontology repo. I am familiar with the ODK schema, but it can be a little hard to follow.
2. I see that OBI uses their edit file for input. I can do the same for PPO now that is has been built, but if one is trying to build an ontology for the first time using only templates, there would be no source for inputs. This is especially problematic if you are creating the ontology's properties in a template and want to use them in another template and run the whole build at one time. In that case, one would have to built a supporting ontology first with the properties.
3. An edge case I know, but what if the property used in a class expression in part of the same template that contains the class expression? Would robot then recognize it as a property?

[ramonawalls]

I updated my ppo-odk.yaml to include

    - filename: PPO_plant_structures.owl
      use_template: TRUE
      template_options: --add-prefixes config/context.json --input https://raw.githubusercontent.com/PlantPhenoOntology/ppo/master/releases/2019-01-16/ppo.owl
      templates:
        - PPO_plant_structures.tsv
     

which generated the following robot command in my makefile

$(COMPONENTSDIR)/PPO_plant_structures.owl: $(TEMPLATEDIR)/PPO_plant_structures.tsv
	if [ $(COMP) = true ] ; then $(ROBOT) template --add-prefixes config/context.json --input https://raw.githubusercontent.com/PlantPhenoOntology/ppo/master/releases/2019-01-16/ppo.owl \
		$(patsubst %, --template %, $^) \
		$(ANNOTATE_CONVERT_FILE); fi

.PRECIOUS: $(COMPONENTSDIR)/PPO_plant_structures.owl

but when I ran make all_components nothing changed in PPO_plant_structures.owl.

[ramonawalls]

I also tried copying the latest release of the file to my src/ontology directory and pointing at it

if [ true = true ] ; then robot --catalog catalog-v001.xml template --add-prefixes config/context.json --input ppo-input.owl \
	 --template ../templates/PPO_plant_structures.tsv \
	annotate --ontology-iri http://purl.obolibrary.org/obo/ppo/components/PPO_plant_structures.owl annotate -V http://purl.obolibrary.org/obo/ppo/releases/2024-07-04/components/PPO_plant_structures.owl --annotation owl:versionInfo 2024-07-04 convert -f ofn --output components/PPO_plant_structures.owl.tmp.owl && mv components/PPO_plant_structures.owl.tmp.owl components/PPO_plant_structures.owl; fi

but it didn't make a difference.

[matentzn]

There are good and bad news.

The bad news, the problem seems real:

robot template --input ontology.owl --template template.tsv -o out.owl

With:

template.tsv:

ID	TYPE	Subclass of
ID	TYPE	SC %
PPO:0001012	class	(PPO:0000010 some PPO:0001004)

ontology.owl:

Prefix(:=<http://purl.obolibrary.org/obo/ppo.owl#>)
Prefix(owl:=<http://www.w3.org/2002/07/owl#>)
Prefix(rdf:=<http://www.w3.org/1999/02/22-rdf-syntax-ns#>)
Prefix(xml:=<http://www.w3.org/XML/1998/namespace>)
Prefix(xsd:=<http://www.w3.org/2001/XMLSchema#>)
Prefix(rdfs:=<http://www.w3.org/2000/01/rdf-schema#>)

Ontology(<http://purl.obolibrary.org/obo/robot-issue-1203.owl>
Declaration(ObjectProperty(<http://purl.obolibrary.org/obo/PPO_0000010>))
)

results in:

<?xml version="1.0"?>
<rdf:RDF xmlns="http://www.w3.org/2002/07/owl#"
     xml:base="http://www.w3.org/2002/07/owl"
     xmlns:owl="http://www.w3.org/2002/07/owl#"
     xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
     xmlns:xml="http://www.w3.org/XML/1998/namespace"
     xmlns:xsd="http://www.w3.org/2001/XMLSchema#"
     xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#">
    <Ontology/>

    <rdfs:Datatype rdf:about="http://purl.obolibrary.org/obo/PPO_0001004"/>
    <DatatypeProperty rdf:about="http://purl.obolibrary.org/obo/PPO_0000010"/>
    
 { REST OMITTED FOR BREVITY }....

It appears that the "oracle" (--input) does not work as intended.

Fortunately there is an easy fix, by simply asserting the type of the object property in the same template:

ID	TYPE	Subclass of
ID	TYPE	SC %
PPO:0000010	object property	
PPO:0001012	class	(PPO:0000010 some PPO:0001004)

It does not matter then wether you are using the --input parameter or not:

robot template --template template.tsv -o out.owl

Works as well.

[ramonawalls]

Thank you, @matentzn! I'll give it a try.

[ramonawalls]

It worked! Woohoo!

[jamesaoverton]

I skimmed this today and recognized the problem. Three hours later I have a partial solution.

The problem is that when the Manchester parser reads "R some X" it doesn't know whether R is a DataProperty or an ObjectProperty. First it tries to get a DataProperty for R, and if that fails then it tries to get an ObjectProperty for R. Maybe there's a reason for that order. For OBO purposes R is more likely to be an ObjectProperty, and it would be nice to be able to look at X to get a hint. But OWLAPI would have to change this, which could be breaking.

The right thing to do is what Nico did: use --input to supply an ontology where the ObjectProperty is declared. His example didn't work because he didn't provide a label for the ObjectProperty. PR #1226 fixes that, at least. (You shouldn't need to put the ObjectProperty in the template.)

I think that's the best we can do.