BEP-0004 - equivalence relationship

[cthoyt]

Closes #6

[ncatlett]

1. I'd suggest the abbreviation = over eq

2. The BEL framework provided for equivalences between namespace values, in which case the assumption that if the same BEL function was used with equivalent values, the BEL terms were equivalent. If I understand correctly, this proposal covers the related need for defining equivalences between BEL terms. A use example that is not handled at the namespace level, but where the BEL entity type is the same is for equivalencing named complexes and composed complexes (in the cases where the named complex represents a specific complex and not a family of specific complexes)

3. Are the BEL terms connected by this equivalence relationship expected to collapse to a single node in a BEL knowledge graph?

4. Any suggestions for how namespace equivalences and term equivalences might be handled within a BEL Framework to avoid potential issues? When resource relationships are included in the main knowledgebase and can be added by any user, there is some potential for clashes or the inadvertent introduction of graph abnormalities?

5. Any thoughts on how to handle equivalences where the BEL entity types are not the same, like the amyloid protein fragment example?

[wshayes]

My expectation is that these equivalences would mostly be added as 'backbone' or automatically generated statements. I would not want to see simple equivalences where the terminologies handle equivalencing, but I do agree that modified abundances need an equivalencing mechanism.

Can you share more thoughts on how to use this for canonicalization (e.g. creating a non-redundant set of edges)? I could see organizing the subject and object so that you chain from subject to object where the ultimate object is the canonical term (e.g. A = B, B = C and C is the preferred canonical term - so adding these statements as B = A while semantically equivalent would not be functionally equivalent. For terms, we decanonicalize to a preferred term for the user based on namespace, but I don't see how to handle that with this construct.

[ncatlett]

Case 4

Another example is handling of "historical" DNA variants where the way that the variant is referred to in the common literature does not match the numbers on reference sequences, e.g., the DNA variant MTHFR C677T SNPedia disambiguates to https://www.snpedia.com/index.php/Rs1801133 BUT dbSNP https://www.ncbi.nlm.nih.gov/projects/SNP/snp_ref.cgi?rs=1801133 does not appear to support disambiguation back to MTHFR C677T

Ideally we'd be able to (1) connect our knowledge graph to a reasonable reference sequence so we can map our data to it (2) enable searching with the name used in the literature - otherwise we may not be able to discover that we have curation around some of these common variants

[cthoyt]

Discusion

Add Case 3: named complex equivalent to composed complex Add Case 4: Genetic Variants in that appear frequently literature that don't correspond to reference sequence. As the reference sequence gets updated, this equivalence becomes problematic. Having this equivalence at the term level and not the namespace is critical.

John: there's a whole system for handling mapping these sites, especially between isoforms and orthologs (different species)

Add strong reccomendation against Case 1:

• Explosion of graph. All of the given identifiers that can be used as a term for HGNC (all resolvable)

John: vote against = since it looks too much like math Natalie: not so strong for = or eq William: no preference

Final vote: eq/equivalentTo

Could we also use sameAs? Probably not the same as OWL since this isn't talking about indiviuals

How do we handle when two equivalent things are different types? Might be different based on the implementation (BEL.bio vs PyBEL) For terms, indicate what term type/entity type? Should we throw a warning?

Natalie: yes William: no, but we should know this somehow Consensus: it's fine for now, but we can define more rules later. We can give some usage guidelines. Use discresion. Maybe we need some guidelines for what equivalence is.

TODO: Needs small update.

[ncatlett]

Probably this comment is too late. Just looking at the long form "equivalentTo" - this sounds directional, where "equivalent" would not be. Is this as intended?