[New metadata element]: mapping_chain_intermediate

[matentzn]

Element id:

mapping_chain_intermediate

Value data type (e.g. URI, URL, text, xsd:boolean):

CURIE (entity)

Example

subject_id	predicate_id	object_id	mapping_justification	mapping_chain_intermediate	comment
MONDO:123	skos:exactMatch	OMIM:111	semapv:MappingChaining	Orphanet:123
MONDO:222	skos:exactMatch	OMIM:321	semapv:MappingChaining	Orphanet:321|SCTID:123	mapping chain can hold multiple values to mean: hoped through multiple intermediates.

Description

Some ontologies such as GO and Mondo curate mappings based on mapping chaining. For example, from:

MONDO:123 --[skos:exactMatch]-->Orphanet:123 and 

Orphanet:123 --[skos:exactMatch]-->OMIM:111

They deduce:

MONDO:123 --[skos:exactMatch]-->OMIM:111

The corresponding justification for this mapping in SSSOM is semapv:MappingChaining. Capturing such chains is very important. In Mondo and GO, mapping_chain_intermediate (in the above example, that would be Orphanet:123) are curated, using the oboInOwl:source property. This issue proposes to adopt this convention as a convenient and easily readable way to capture chain metdata.

Additional thoughts

I have been convinced through various discussions that this property is convenient, easy to understand and informative. However, for the sake of completeness I want to reveal that my preferred way of modelling this would have been to explicitly model mapping provenance by documenting the full path MONDO:123 --[skos:exactMatch]-->Orphanet:123--[skos:exactMatch]-->OMIM:111 in some way. Documenting the path as a string is a bit verbose (MONDO:123-skos:exactMatch->Orphanet:123-skos:exactMatch->OMIM:111 or whatever) and the more elegant way, referring to a mapping directly by identifier, is not possible because we decided to no have mapping_id in SSSOM so far (a globally unique identifier scheme for mappings would provide a huge burden to global integration and dependency on infrastructure which would make sssom much more heavyweight).

[cthoyt]

I don't think it's clear what kind of thing would go in this field from what's written currently

[matentzn]

OOps, yes. Updated with an example.

[cthoyt]

What if the predicates are different through the hops?

[matentzn]

Yes, this is exactly my argument againt this proposal. The predicates are collapsed. What this is saying is this:

"We traced a mapping chain through A,B,C and determined that this is sufficient to justify the current mapping".

Some information is lost, but it is maybe justifiable for the tersity gained. intermediates are very valuable evidence regardless of whether the mapping predicate was respected. Having the predicated would, however, make it even more explicit - at the cost of simplicity and readability.