Bottom-up discovery of inverse properties

[mereolog]

There is a couple of triples at the individual level that exhibit inverse properties that are not modelled at the class level - see the query below.

Consider this example:
Norvasc5mgTabletConstituency hasActiveIngredientRole AmlodipineBesylateAsActiveIngredientInNorvasc5mgTablet.
AmlodipineBesylateAsActiveIngredientInNorvasc5mgTablet isRealizedIn Norvasc5mgTabletConstituency .
At the same time the respective properties are not inverses of one another - even if you include their superproperties.
This may indicate

1. either (at least) one of these triples is wrong
2. or we can say in the ontology that a super property of hasActiveIngredientRole is an inverse of isRealizedIn or something equivalent.

Other cases like this one are listed in
inferrable inverses.csv.

SPARQL :

PREFIX owl: <http://www.w3.org/2002/07/owl#>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT DISTINCT *
{
    ?r1 ?p1 ?r2.
    ?r2 ?p2 ?r1.
    FILTER (REGEX(STR(?r1), 'pistoia|omg'))
    FILTER(ISIRI(?r1))
    FILTER(ISIRI(?r2))
    FILTER NOT EXISTS {?p1 (rdfs:subPropertyOf*)/(owl:inverseOf|^owl:inverseOf) ?p2}
    FILTER NOT EXISTS {?p1 (owl:inverseOf|^owl:inverseOf)/(rdfs:subPropertyOf*) ?p2}
    FILTER NOT EXISTS {?p2 (owl:inverseOf|^owl:inverseOf)/(rdfs:subPropertyOf*) ?p1}
    FILTER NOT EXISTS {?p2 (rdfs:subPropertyOf*)/(owl:inverseOf|^owl:inverseOf) ?p1}
}

[ElisaKendall]

@merelog It's possible that we could add a few more restrictions, but some of these are related to property chains. isRealized in, for example, is a property chain. I'm seeing a few things that I want to address, but may need to talk with you about how to address the chains in these cases.

[ElisaKendall]

@merelog I've addressed the issue with quantities and units, but because of the property chains prefer to pause on the active ingredient role related issues this raises for the time being. We can come back to this in the next phase of the project as appropriate.