Gracefully handle synonymous curie prefixes using the curies library

[kevinschaper]

Related to some discussion about handling RXNORM vs RXCUI, NN could provide prefix normalization along with node normalization very easily by making use of the curies library and the merged prefix map from prefixmaps.

@matentzn and I would love to help out on this.

The code is as easy as:

>>> from prefixmaps import load_converter
>>> converter = load_converter("merged")
>>> print(converter.standardize_curie("ORPHA:123"))
ORPHANET:123

Though, there is something to dig into, because I'm finding that converter.standardize_curie("RXNORM:123") isn't returning RXCUI:123 like I was expecting. Which, uh, I recognize weakens the argument a little! 😬

[gaurav]

I don't think NodeNorm should be messing with the CURIEs it gets from Babel -- there's too much of a risk that e.g. ORPHA:123 is normalized to ORPHANET:123, but we already have ORPHANET:123 in a different clique for some reason. Especially if this is relatively easy to do with the NodeNorm output -- I'm not sure there's a big benefit to incorporating it into NodeNorm, and having an additional bit of provenance to content with (i.e. which version of which prefixmap was used to normalized CURIEs).

I do wonder if this is something we could incorporate into Babel -- we would have use Biolink Model's prefixmap to keep in sync with Translator, and something we're fussy about is making sure that cliques have the identifiers from NodeNorm in the order that Biolink Model has its valid ID prefixes (e.g. NCBIGene, ENSEMBL, etc. for biolink:Gene). At a minimum we could produce a report of all the identifiers that have the "wrong" normalized CURIE, but other possibilities would be adding the normalized CURIE to the list of identifiers as long as they are also included in the valid ID prefixes (e.g. you might have both ORPHA:123 and ORPHANET:123 as IDs in a particular clique). Would something like that help your issues?

[matentzn]

there's too much of a risk that e.g. ORPHA:123 is normalized to ORPHANET:123, but we already have ORPHANET:123 in a different clique for some reason.

This is not ideal though - There is not ORPHANET:123, they should both be the same. We have detailed databases (also implemented in curies) for prefix synonyms, and its worth taking a look it, even if there are say 0.1% false positives.

I think having at least on optional prefix normalisation step would help massively with recall. Prefix normalisation should conceptually be part of node normalisation. If someone wants to normalise RXNORM:123 right now gets no response, despite RXNORM being a very well known synonym of RXCUI (which NN understands).

I understand that the effort may be big, but for now: with infinite funding, would you be happy to have this part of NN, even if it meant that you might get a few more cliques that would need manually resolution? (remember, infinite funding.. :P )