ChEBI subsets

[joeflack4]

So far creates sub-hierarchy of ChEBI for only what is mapped to LOINC. Main future goal is to use to create alt LOINC hierarchy.

Changes

ChEBI subsets

• Add: makefile: To add goals for creating these outputs.
• Update: .gitignore: To include folders for these inputs/outputs.

Results

Google Drive

[joeflack4]

Where to put ChEBI inputs

I have the goal of moving all of the sources somewhere in input/. I've decided that'll probably be input/sources/, or something to denote all of the inputs that actually go in to generating our outputs.

As I'm not sure yet if ChEBI will be such an input, I'm putting it in input/analysis/.

I also plan to eventually move owl-files/ and data/ into input/, but won't do that in this PR.

[joeflack4]

makefile

@ShahimEssaid The makefile is back. I find that this will likely be the best way to do this work. Let me know if you have any thoughts otherwise.

Added an all goal which just includes chebi-subsets.

chebi-subsets includes:

CHEBI_OUT_BOT=output/analysis/chebi-subset-BOT.owl
CHEBI_OUT_MIREOT=output/analysis/chebi-subset-MIREOT.owl

These goals require this as input:

CHEBI_MODULE=output/analysis/chebi_module.txt

...which queries:

PART_MAPPINGS=loinc_release/Loinc_2.78/AccessoryFiles/PartFile/PartRelatedCodeMapping.csv

Results: Google Drive

[matentzn]

Note there are more mappings to be found here: monarch-initiative/monarch-mapping-commons#35

[joeflack4]

Extracting module via awk

@ShahimEssaid For extracting the unique list of mapped ChEBI terms from the LOINC mapping CSV, I wanted to not rely on Python for this, but maybe I'll change that for a few reasons:

1. Windows doesn't come with awk, etc.
2. Couldn't get it to display commented labels next to the terms.
3. The awk command, and parsing CSV with default unix commands, is still non-trivial. It's hard to read / maintain.
4. Has an off-by-1 (row) error.

[matentzn]

If its easy enough, write a sssom adapter and add to sssom-py? Note that you will have other mappings to process, e.g. monarch-initiative/monarch-mapping-commons#35, since the LOINC part mappings is always incomplete, so makes sense to try and convert everything to sssom first

[joeflack4]

Hmm, I think the SSSOM part is a separate issue, but thanks for the reminder. I think for this proof of concept a very quick pandas func will suffice.

Are there some docs on what it means to create a "SSSOM adapter" for a given source?
I wonder if that may be non-ideal / impossible in this case, because the source mapping CSV requires download of the full LOINC release, which is also behind a license.

[joeflack4]

How will we use these outputs?

One/both of these?

• Grouping classes: from ChEBI #55
• ChEBI: Alternative hierarchies #108

Or do we have other goals in mind?

[matentzn]

Ontology PURL

This is more a general thing but always use PURLs for ontology download locations

[joeflack4]

@matentzn This is a good idea. Don't we typically use Bioregistry as the central location for getting canonical PURLs?

• https://bioregistry.io/registry/chebi

However, I don't see an .owl URI there, only URIs for prefix maps.

There is a URI for it at Ontobee:

• https://ontobee.org/ontology/CHEBI

These PURLs have the disadvantage though of pointing to the .owl, not the .owl.gz, which I think is a better option whenever it is available.

[matentzn]

No, we should always use official OBO PURLS:

http://purl.obolibrary.org/obo/chebi.owl
http://purl.obolibrary.org/obo/chebi/chebi.owl.gz

etc..

[joeflack4]

@matentzn OK that's good. I just pushed a commit and now it's using the PURL!

---

But what I'm sort of asking is how do I know what the PURL URI is for a given ontology URI? Is there somewhere that I can search?

I thought that place was BioRegistry, but that's not the case here.

I suppose I could just type out http://purl.obolibrary.org/obo/MAIN_ONTOLOGY_SPELLING.owl or http://purl.obolibrary.org/obo/MAIN_ONTOLOGY_SPELLING/MAIN_ONTOLOGY_SPELLING.FILE_EXTENSION and check to see if they exist, but that's not a good UX.