Inflammation: linking process, disease and phenotype terms in EFO

[paolaroncaglia]

In this ticket I'll clean up and update the discussion (started in #372) on how to represent inflammation in EFO, including suggestions on how to avoid duplication and how to link the biological process of inflammatory response, the phenotype of inflammation, and the disease branch of inflammatory diseases. We may consider this a "poster child" for the EFO Epic on "Interoperable disease/phenotype design patterns" (EBISPOT/ontology-webpages#10).

Wrt biological process:

EFO has GO_0006954 'inflammatory response' (http://purl.obolibrary.org/obo/GO_0006954)
This is a GO Biological Process term (child of ‘response to stimulus’) and is in the EFO process branch. It has the gwas_trait true annotation, meaning it’s being used in GWAS, so we’d like to retain it. It has 2 children (1 with further descendants). These are general terms, not related to specific diseases/disorders. However, this GO term is currently used in EFO in axioms of some disease and measurement terms, in the following ways:
disease X (e.g. aortitis) participates in some inflammatory response;
disease X inheres in some inflammatory response;
measurement X (e.g. basophil count) is about some inflammatory response.

• Make a summary spreadhseet of the 114 class usage cases of 'inflammatory response' in EFO3;
• Evaluate if changes should be made.

MONDO uses the same GO term as above in the following axiom:
MONDO_0021166 'inflammatory disease' RO_0004024 'disease disrupts' some GO_0006954 'inflammatory response'

• Note, this is not visible in OLS as of today 20/6/19 (OLS not showing full axioms (nor synonym metadata) OLS#229 (comment))
• Note, the RO property is called 'disease causes disruption of' in RO but 'disease disrupts' in MONDO
• The RO property above is not in EFO3 (as of today 20/6/19)

I propose to have the same in EFO, i.e.:

• Import RO_0004024;
• Add
  EFO 'inflammatory disease' RO_0004024 'disease disrupts' some GO_0006954 'inflammatory response'
• Clean up "inappropriate" usage of GO_0006954 'inflammatory response' in EFO, presumably by replacing with an axiom to the phenotype term of choice for inflammation, as discussed further on.
  Benefits: improved precision in axioms; enrichment of information content; alignment with MONDO.
  Reminder: this link to biological process will not exclude or replace links to phenotypes. Those will be addressed next.

To be continued.

Very personal note: this is one of the deepest rabbit holes I've come across, and I've been a GO editor for 6 years...

[paolaroncaglia]

Continued:

Wrt phenotype:

Based on previous recommendations (see #372) and on current view of HP, that (now?) has a term in HP-proper space to represent phenotypes of inflammation AND that links to the GO term above:

• Add link: EFO 'inflammatory disease' 'disease_has_feature' some HP:0012647 'Abnormal inflammatory response'
• Coordinate with MONDO, check if they wish to do the same
• Before the end of this project, clean up the branch under EFO HP:0012647 'Abnormal inflammatory response' if needed

Based on previous recommendations (see #372), do not obsolete
EFO MP inflammation (http://purl.obolibrary.org/obo/MP_0001845)
and
EFO MPATH inflammation (MPATH) (http://purl.obolibrary.org/obo/MPATH_212)
but

• tag them so people don't use them any longer (e.g. add annotation property of editor_note to say that terms are no longer valid, and indicating what inflammation term should be used instead).
• clean up descendants and/or usage cases for both

To be further continued...

[paolaroncaglia]

And finally to wrap up,

Wrt object properties and axioms:
To clean up inflammatory diseases scattered in EFO and establish firm guidelines going forward:

• EFO has EFO:0000784 'has disease location', while MONDO has RO:0004026 'disease has location'. Should we replace the EFO property with the RO one, to improve interoperability? Infact, at our EFO meeting on 22/11/2018, we thought that we may consider that (as recorded in Disease location object property #312). And we already previously started on a plan to clean up non-RO object properties in EFO where applicable (Cleaning up EFO object properties #21).
• non-inflammatory diseases should have axioms 'has disease location/disease has location' when location site is known and is unique
• inflammatory diseases should have axioms 'disease has inflammation site' when inflammation site is known and is unique
• 'disease has inflammation site' is a child of 'has disease location' (or should become a child of 'disease has location' if that becomes the property of choice) - would that help to automate inferred reclassification? Note we wouldn't want to lose asserted links to other diseases.

[paolaroncaglia]

Lastly,

• Inform users of the changes, especially GWAS (bcs they use the GO term in EFO). Do this in EFO release notes + email.

If I've forgotten anything, please add below. Thanks.

[paolaroncaglia]

Tagging the main relevant people, though feedback is welcome from everyone of course:
@cmungall @nicolevasilevsky @pnrobinson
@simonjupp @zoependlington @matentzn @dosumis
Could you please review my proposal to represent inflammation in EFO and approve or object. This is a big ticket; I tried to keep things to the point, so if you have comments that go beyond this strict topic, please open new tickets. Many thanks in advance for your feedback.

[nicolevasilevsky]

Biological process
Your plan for biological process sounds fine to me.
I'm not sure why RO_0004024 has a different label in Mondo.
Personal note - yes, this is very complicated! Thanks for your efforts to address this.

phenotype
I like the idea of adding 'disease_has_feature' some HP:0012647 'Abnormal inflammatory response' to 'inflammatory disease' in Mondo. Chris should confirm before we implement this.

object properties and axioms

• I am in favor of EFO aligning with Mondo wrt to the relations
• I think your other proposals sounds good, I don't see any issues with them

Thanks @paolaroncaglia for all your work on this!

[paolaroncaglia]

Quick notes from the EFO-MONDO meeting 24/6/19:
Nico: ok to go ahead with disease X 'disease has feature' phenotype Y, if it's only used in a handful of cases in EFO; be aware that it may change to phenotype X 'is underlying phenotype' of disease Y or similar in the future.
Dani: ok with everything except for EFO 'inflammatory disease' RO_0004024 'disease disrupts' some GO_0006954 'inflammatory response'; Paola: on the other hand, if we implement the rest of the plan, we wouldn't need that axiom to link to the GO term.

[matentzn]

Note that disease has feature is more general than ‘has defining feature. I feel that these relationships are so important that should be curated in the Mondo repo rather than here. But let’s start small and document well! :) Ideally, use csv files rather than editing in protege for such mappings.

[paolaroncaglia]

@cmungall
I know this is a long thread, but when you have a chance, could you please review and comment?
Ideally, I'd like to implement the plan before the next EFO release.
Many thanks,
Paola

[paolaroncaglia]

@zoependlington
Any chance you could help with the following please, I think you mentioned you had started to look into it?

• Make a summary spreadhseet of the ~114 class usage cases of 'inflammatory response' in EFO3
  Thanks! :-)

[paolaroncaglia]

A quick note for self that, as part of this 'inflammation' ticket, we'll need to look into the following duplications in EFO:

EFO > MONDO:0002102 cheilitis (subclass of inflammatory disease and lip disease)
EFO > HP:0100825 Cheilitis (subclass of inflammation)
Note that the MONDO term xrefs the HP term.

EFO has
EFO > HP:0012115 hepatitis (a phenotype)
EFO > EFO:0004196 viral human hepatitis infection (a disease, with 4 children)
MONDO has MONDO:0002251 hepatitis with child MONDO:0006011 viral hepatitis)