Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add "disease or disorder" to COB #226

Draft
wants to merge 3 commits into
base: master
Choose a base branch
from
Draft

Add "disease or disorder" to COB #226

wants to merge 3 commits into from

Conversation

matentzn
Copy link
Contributor

Addresses #19

Ludi Incipiant.

@wdduncan
Copy link
Member

Oh boy ... let the games begin :)
@matentzn What would the definition be for disease or disorder?

src/ontology/cob-edit.owl Outdated Show resolved Hide resolved
@matentzn
Copy link
Contributor Author

:) I added it to the PR! https://github.com/OBOFoundry/COB/pull/226/files#r1088050803

@lschriml
Copy link

Question: Is the plan to add a term in COB as:
'disease or disorder' ??

I would request to have disease added as a it's own COB term.

And the term disorder, if other want it, as it's own distinct term.

Thank you,
Lynn

@wdduncan
Copy link
Member

wdduncan commented Jan 26, 2023

@matentzn I think the definition you reference is the source of contention:

A disease is a disposition to undergo pathological processes that exists in an organism because of one or more disorders in that organism.

This is a definition for disease and begs questions about what is a disorder.

A more fruitful approach may be to follow @cmungall suggestion to adopt Schultz' pathological structure.

I think it is also reasonable to stipulate that COB uses the term 'disease' in the dispositional sense of the word, and create other terms for other senses of the 'disease' (e.g., disorder).

In any case, I fear this conversation will never end.

@addiehl
Copy link

addiehl commented Jan 26, 2023

+1 to Bill's comment. Adding a single class labeled 'disease or disorder' seems to be straight out favoritism of the MONDO approach and ignores the distinctions between disease and disorder used by OGMS and related ontologies like the Infectious Disease Ontology and DO. Might be better to keep 'disease', 'disorder', and 'disease or disorder' out of COB rather than alienate folks using alternative approaches, particularly if COB becomes a requirement for OBO Foundry membership.

@hoganwr
Copy link

hoganwr commented Jan 26, 2023 via email

Copy link

@lschriml lschriml left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Consesus of usage is to use the label: disease

Defintion: I am open to further review/revision.

Usage of the term 'disorder', in disease nomenclature, has often been used across research/clinical communities interchangeably with disease, e.g. mental health disorder, platelet-type bleeding disorder, intellectual developmental disorder.

To address, what does the 'concept' of 'disorder' mean:
Clinically, a disorder precedes a disease, and is a congenital condition.
That is to say, it is present at birth.

In the DO we define 'physical disorder' as : A disease that has_material_basis_in a genetic abnormality, error with embryonic development, infection or compromised intrauterine environment.

src/ontology/cob-edit.owl Outdated Show resolved Hide resolved
src/ontology/cob-edit.owl Outdated Show resolved Hide resolved
src/ontology/cob-edit.owl Outdated Show resolved Hide resolved
@lschriml
Copy link

lschriml commented Jan 26, 2023 via email

@lschriml
Copy link

lschriml commented Jan 26, 2023 via email

@matentzn
Copy link
Contributor Author

matentzn commented Jan 26, 2023

"Mondo favouritism" is a great accusation to start a discussion with if you want it to finish quickly!

In Mondo, it was decided to treat disease and disorder synonymous - which means Mondo classes will never be aligned with COB if we separate the two. COB is explicitly not supposed to be BFO. The point is to create categories that are useful for biology / biomedical domain. Schulz conflation won't solve the disease vs disorder debate.

The matter of fact is:

mondo, ncit, ogms, doid and all other interpretations of "disease" can go under "diease or disorder". Mondo and NCIT cant align with a dual "disease" ... "disorder" solution.

EDIT: I am NOT the right person to have this discussion. I only care about one thing: the success of OBO and the unification of OBO ontologies.

@lschriml
Copy link

lschriml commented Jan 26, 2023 via email

@alanruttenberg
Copy link
Member

If "disorder" is used as a synonym for disease, then add "disorder" as an alternate term. I believe the objection is that it's important to keep a distinction between disease and disorder, whatever they are called. Offering the disjunctive class will encourage its use, and then uses of it will mean one can't make the distinction.

@matentzn
Copy link
Contributor Author

I don't see how doid and all other interpretations of "disease" can go
under "disease or disorder

Can you elaborate @lschriml? the idea here is to create a conjunctive class that conflates the concepts of disease or disorder. Disease is a subclass of Disease or Disorder, so is "Disorder". This term can encapsulate all interpretations of disease/disorder, including "disease as a disposition" or "disease as a material entity". In any case, remember the point of COB: have a useful upper level - not a BFO conformant one. Now here is the crux I think: we may have different definitions of useful. For me it is: integrate all OBO ontologies into a reasonable knowledge graph that can be used, for example, to group genetic mechanisms for disease, or phenotypic profiles associated with them and provide background knowledge for AI systems of the future. This is a large scale, pretty dirty affair - unless you believe that everyone curating this kind of data (GWAS?) is thinking about the fine distinctions between disease and disorder. Or worse NLP: how will an NLP process ever understand whether a paper is talking about a disorder or a disease? Reality (biocuration, NLP) is dirty - if we insist cleaning it up BFO-style, we will never integrate our data, we will never build the grand OBO biomedical KG, and we will slowly fade into irrelevance, seeding ever more of our hard-earned knowledge into the abyss of neural networks with 1000 of layers and no way for us (as humanity) to even begin to grasp what's happening. It's our role as OBO to represent the Human side of the upcoming AI revolution, and hackling about fine-grained ontological distinctions will simply prevent us from being integrated into the AI stack that is being built right now.

@lschriml
Copy link

lschriml commented Jan 27, 2023 via email

@zhengj2007
Copy link

+1 to Bill and Alex

I am OK with Alan's recommendation: add "disorder" as an alternate term to 'disease'

@matentzn
Copy link
Contributor Author

Disorder is a type of disease? I guess if that is true, than we can solve this problem here by just renaming the class to disease and just getting common understanding that the disease class also subsumes all disorders? That could be doable.

@alanruttenberg
Copy link
Member

@matentzn The suggestion isn't that disorder is a kind of disease. The definitions are primary. If someone wants to use the word disorder to mean what the disease definition says, that's ok, and that's what my proposal was about. There would still be a class disorder whose primary label will remain disorder. It would be a mistake to put a bona fide disorder as defined as a subclass of disease.

@matentzn
Copy link
Contributor Author

While I disagree, can we make a suggestion that allows us to define a single term that encapsulates all ontologies including DO, Mondo and NCIT? Something sane we can use as a parent class.

@wdduncan
Copy link
Member

I think it would really helpful if a definition can be provided for disease or disorder that permits the conflation you wish. At present, the definition doesn't allow this. We are only relying on our intuitions about what kinds of things would be encompassed by this conflation.

@wdduncan
Copy link
Member

wdduncan commented Jan 27, 2023

While I disagree, can we make a suggestion that allows us to define a single term that encapsulates all ontologies including DO, Mondo and NCIT? Something sane we can use as a parent class.

@matentzn This is what I am asking for. A definition for such a term has yet to be provided.

@wdduncan
Copy link
Member

FWIW: NCIT also has a top-level Disease, Disorder, or Finding class. Do you wish to conflate findings too?

@matentzn
Copy link
Contributor Author

You know me @wdduncan i have zero interest in defining things and a huge passion for integrating things that are defined in incompatible ways - I don't know what I want other than to start using COB and putting a domain on “disease has feature” relationship in RO (and all the other disease relations). I neither know what a disease is, not what a finding is; I just need a class to do my work of building the OBO knowledge graph.

@nataled
Copy link

nataled commented Jan 27, 2023

Clinical abnormality as term name? I'll leave the definition to the experts.

@wdduncan
Copy link
Member

wdduncan commented Jan 27, 2023

@matentzn Unfortunately, I think in order of this proposal move forward (or gain consensus) a definition needs to be provided. As far as I know, giving such definitions has been a principle for the Foundry since its inception.

@cstoeckert
Copy link

I'd like us to consider the approach raised by @cmungall described in Schulz et al. Journal of Biomedical Semantics 2011. @matentzn indicated above that the “Schulz conflation’ won’t work but it's not clear to me why it wouldn't. If that approach works it has the.benefit of the work done to relate to SNOMED (described in the paper) and potentially offer an approach for addressing phenotypic abnormality and perhaps phenotypes in general.
Drawing from the Results section of the paper and starting with COB classes we could have:
COB:pathological entity mapped to bt:pathological entity
COB:pathological entity equivalent to COB:pathological structure or COB:pathological disposition or COB:pathological process
COB:pathological structure mapped to bt:pathological structure
COB:pathological structure mapped to OGMS:disorder
COB:pathological disposition mapped to bt:pathological disposition
COB:pathological disposition mapped to DOID:disease
COB:pathological disposition mapped to MONDO:disease or disorder
COB:pathological process mapped to bt:pathological process
We can discuss whether any of the mappings are equivalent.

@lschriml
Copy link

lschriml commented Feb 2, 2023

In order to make progress on COB, could we all agree to add the two terms (1) disease and (2) disorder to COB.
And to continue the other discussions, points raised in this issue, for a later date/time.

@zhengj2007
Copy link

In order to make progress on COB, could we all agree to add the two terms (1) disease and (2) disorder to COB. And to continue the other discussions, points raised in this issue, for a later date/time.

It sounds good to me.

@matentzn
Copy link
Contributor Author

matentzn commented Feb 3, 2023

I am against doing this, sorry. It means creating a solution that does not work for Mondo and make it impossible to negotiate a solution later on that does. Sorry!

@alanruttenberg
Copy link
Member

I'm very confused. As I pointed out, MONDO only /calls/ the term 'disease or disorder'. The definition is that of disease as given in OGMS. Adding 'disease or disorder', defined as 'disease' or 'disorder' doesn't help MONDO AFAICT.

@matentzn
Copy link
Contributor Author

matentzn commented Feb 3, 2023

The definition does not reflect practice and I asked the Mondo team to remove or revise it. it does not reflect the current interpretation. Mondo does not distinguish between disease or disorder, the two concepts are conflated in the Mondo view of the world! They even met this week with a few medical experts to discuss this issue and the decision remains: while a small difference between disease and disorder seems to have been acknowledged conceptually, for pragmatic reason they are treated as synymous/conflated.

@alanruttenberg
Copy link
Member

In that, case, once the definition is revised, which should be preliminary, the presence of disease and disorder in COB will allow then to completely define disease or disorder in MONDO. There is no need to have the disjunction in COB. It's a defined class.

@matentzn
Copy link
Contributor Author

matentzn commented Feb 3, 2023

I had 5 people sending me emails and messages to recommend to change a bit how I communicate here, but right now, I need time to retreat and formulate a new way to argue this case. I will have to look also at @cstoeckert and @addiehl proposals a bit more. Also, it seems like my position is not understood at all (what you are suggesting does not solve my problem - I need 1 named class to be able to query all diseases and disorders at once), so I probably need to make some pictures and organise an in-person meeting to solve this issue! I cannot agree to adding the strict disease/disorder distinctions proposed here.

@alanruttenberg
Copy link
Member

Ok. Regroup. I'm game when you've organized your thoughts.

@matentzn
Copy link
Contributor Author

matentzn commented Feb 3, 2023

Argument for a named class that captures Mondo and DO definitions.

Note: This is not an argument for or against any particular definition, just establishing the need for a named class that captures the Mondo and DO viewpoints. We still need to think about how best to achieve a minimal model that makes everyone happy.

Just because I am at the airport, let's start with my motivation, not the modelling now.

Use case 1: querying

How would you write this query without a sane grouping?

image

Use case 2: Sane domains and ranges

I am assuming here we don't want all our relations to have a disjunctive expression as a domain, because this will cause problems for lower expressivity ontologies, and create a big overhead maintaining properties.

image

Use case 3: COB alignment as a strategic goal of the OBO Foundry

Using disjunctive expressions for alignment purposes will simply make the purpose of this effort nil.

image

@wdduncan
Copy link
Member

wdduncan commented Feb 3, 2023

For what it is worth, I wish we would explore the so-called Schultz approach (mentioned multiple times) a bit more instead of going back and forth endlessly about disease/disorder. Or, we should use different labels so that we don't continue arguing about which senses of 'disease' or 'disorder' are correct, and whether it is best to conflate the two.

But since those options have not garnered a lot of interest, I think the best option is to logically define disease or disorder as the union of disease or disorder, where disease is subtype of disposition and disorder is a subtype of material entity. This allows MONDO to use this class as well other ontologies that need the disposition / material entity distinction.

I realize there are drawbacks to this:

  1. It may encourage annotators to gloss over relevant/important distinctions between the dispositional and material senses of 'disease'
  2. To quote from above: "There is nothing to stop anyone to align with a huge disjunction of all class in COB."

As for (1), there are (based on @matentzn comments) are a fair number of professionals who are not bothered by the conflation. We should respect that. I would, however, be concerned if more of the conflated subclasses from MONDO wanted to be entered into COB. This could impose the conflation between disease and disorder on those of us who need to clearly disambiguate between diseases (as dispositions) and disorders (as material entities). I have worked with researchers in the domains of cancer and dentistry that were quite concerned with being clear about these distinctions.

As for (2), I'm not seeing why this is an alarming concern. Sure ... it is possible, but whether it actually happens is a different matter. There is nothing to stop other OBO ontologies from doing this within their own domains, but we don't find them.

@nataled
Copy link

nataled commented Feb 3, 2023

The label aspect is one I thought might be the problem, hence why I suggested a different one.

@bpeters42
Copy link
Contributor

For what it is worth: I am with Chris Stoeckert spelled out comment on how to re-label and align these 'pathological classes' based on the Schultz approach (It is SCHULTZ, not SHULTZ - we are talking real Germans here). That approach also allows people like myself that want to try to fit specific terms into the lowe level buckets as much as possible. But gives room to integrate MONDO.

@wdduncan
Copy link
Member

wdduncan commented Feb 8, 2023

@bpeters42

It is SCHULTZ, not SHULTZ - we are talking real Germans here

Yes. Thanks for pointing that out ... I corrected my misspellings.

I noted the Schultz approach (here) as an avenue to explore (not necessarily to commit to) for finding a middle ground between OGMS/DO and MONDO positions. While I am happy to see that @cstoeckert and @bpeters42 support this, I fear that the proposal to map COB to Biotop classes will (yet) create more arguments. This is something that needs to be considered as well.

To be honest, I don't see a perfect solution here. Each side has advocates that are unwilling to budge.
I endorsed the logical disjunction of disease or disorder in hopes that each side would see that are getting enough of what it wants to tolerate the other view points. A kind of ontological détente (I suppose).

Unfortunately, I fear my efforts have failed. I am planning to attend to disease/disorder call. Until each side starts seriously thinking about solutions that don't require jettisoning the other, I have strong doubts that anything will result from the call. In other words, if the call only amounts to each side saying why they are right and the other side is wrong, this debate will just continue.

@alanruttenberg
Copy link
Member

I think the real problem here is the overlap of MONDO and DO and lack of a unified plan for how the Foundry will deal with pathology. Objection to the change stems from the feeling that we would be cementing in the current dysfunction with no path towards resolving the elephant in the room. I don't think anyone thinks that the Schulz/OGMS distinctions, sans disjunction, are bad distinctions or difficult to distinguish. I think that I would not have objections to the addition of the disjunctive term if there was a commitment to a) resolve the redundancy between DO and MONDO and b) commit to, in the longer term, refining terms to be in the more specific class. If there's some way to shift the conversation in that direction, we should attempt it. If you agree this is a plausible way forward, perhaps we could craft a message together to shift the conversation.

Absent that, I don't think I would support going forward with the change, and would instead modify the dictum that every ontology had terms directly below a COB term to be instead that every ontology be under a term fully defined in terms of the COB terms. That would allow for disease or disorder to still be conformant, but stay in MONDO. In speaking with Barry it was his suggestion that regardless of this issue, there should be the ability root under defined classes based on the COB terms.

@nataled
Copy link

nataled commented Feb 23, 2023

My recollection is that early discussions touched on the issue of 'this or that' (defined class) terms, and that it was decided they would NOT be in COB. This of course would not work for those ontologies that make use of such terms at the very top levels but, again, this was previously viewed as okay.

@ddooley
Copy link
Contributor

ddooley commented Mar 23, 2023

Where can I catch up on the result of the special meeting on "disease or disorder"?

@bpeters42
Copy link
Contributor

bpeters42 commented Mar 26, 2023 via email

@matentzn
Copy link
Contributor Author

I am afraid someone will need to dedicate some serious time to shepherd this issue forward. One option is to recruit a volunteer to OBO Ops whose sole responsibility is to mediate a solution, summarise the current situation from time to time and make a strategy?

@matentzn matentzn marked this pull request as draft April 28, 2023 12:40
@bpeters42
Copy link
Contributor

Would you have a volunteer in mind? If so, I would love that solution.

@matentzn
Copy link
Contributor Author

matentzn commented May 4, 2023

There are a few people I could imagine doing this (Role: COB mediator - a person whose role is the mediation of modelling disputes on the COB issue tracker). The problem is that the most suitable people are the ones that already have a very full plate. For me the key ingredients of a mediator are:

  • reasonableness and not too strong personal opinions (excludes everyone that has already commented on this thread)
  • ability for organising information (in this case, collecting and presenting arguments)
  • sociability in the sense that they can work with people without riling them up
  • a thick skin and proactivity to drive difficult discussions forward.
  • Enough understanding of ontologies to understand the "building a formal ontology BFO style vs creating a semantically uncommitted upper structure for data integration" dichotomy that we have

A few people I think of that would fit the bill above:

  • Allen Baron (DO team, of course I don't want to throw him to the wolves, but he is just very reasonable through and through)
  • Sierra Moxon (Monarch/NCATS translator)
  • Tiago Lubiana (Postdoc, Wikidata, Cell Ontology)
  • Jim Balhoff (RENCI)

We could reach out to them 1 by 1, or we could propose the role on OBO level and see if people volunteer that I have not thought of yet.

@alanruttenberg
Copy link
Member

What does it mean for an ontology to be "semantically uncommitted"?

@matentzn
Copy link
Contributor Author

matentzn commented May 4, 2023

All root classes are subclasses of BFO:entity seems quite uncommitted wrt TLO, compared to using very specific BFO distinctions. The same logic would apply to COB.

@bpeters42
Copy link
Contributor

I just wanted to add to the comment from Nico regarding semantic commitment: Most high-level COB classes are BFO classes or children thereof. We have some problem cases (like here: disease/disorder) where there is disagreement on the placement. But we are working towards trying to resolve that. In the meantime, we accept these uncommitted classes in order to make OBO ontologies interoperable as best we can - which is the primary goal for COB.

@allenbaron
Copy link

I appreciate the desire to add disease and disorder to COB. I didn't attend the last meeting on this (April?) but I was under the impression that it had been decided in that meeting that these two terms would be added (using the OGMS definitions? not sure about specifics).

What else, if anything, needs to be done?

I apologize for the long delay between this and previous responses.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.