Babysteps towards sorting out the Phenotype/Disease divide

[souzadevinicius]

To start sorting out the disease vs phenotype business, we suggest the following plan:

1. Update our HPO - Mondo mapping by running the OAK lexmatch matching pipeline to HPO.
2. Implement a review process that separates the following cases:
   1. This is a genuine phenotype and should not be a disease (MONDO ID should be deprecated)
   2. This is a genuine disease and should not be a phenotype (HPO ID should be deprecated)
   3. The Label is used in practice to refer to either a disease and a phenotype, but they are clearly conceptually different
   4. The Label is used in practice to refer to either a disease and a phenotype, but they are not clearly conceptually distinct
   5. (This is neither a disease nor a phenotype, both IDs should be deprecated)
   6. Develop a strategies for:
      1. Case 3 (clearly conceptually different):
         1. Solution 1.1: Leave as is and say "conceptual disambiguation are what ontologies there for". Document. Share documents. Run trainings.
         2. Solution 1.2: Introduce a unique naming scheme (e.g. put phenotype or disease in brackets after the LABEL for HPO and Mondo releases.
         3. Independent of the above, consider if the two can be linked ontologically ('has defining feature' or some such)
      2. Case 4: Both appear to be conceptually the same, but it is unclear if they are a phenotype or a disease.
         1. Solution 1: We firmly establish once and for all the phenotypes and diseases are conceptually disjoint, and force the matter.
         2. Solution 2: We define a new model of disease phenotype that is more permissive and allows for "conflation"

Steps for now:

• [x] @souzadevinicius to run oak lexmatch to match HPO and Mondo and update Mondo-HPO mapping
• [x] @souzadevinicius to create a Google sheets with these columns: Google Spreadsheet

mondo_id	mondo_label	hpo_id	hpo_label	case	justification	reviewer
MONDO:123	mondo label	HP:123	hpo label	Mondo ID should be deprecated (Case 1)	Because...	Nico

• [x] @pnrobinson Document our model of disease vs phenotype for curators to comprehend (being refined)
• [ ] https://github.com/monarch-initiative/mondo/issues/7494
• [ ] Run a test curation to see what comes out. The best outcome would be that most cases are between Case 1, 2, 3, 5. These all have reasonably easy solutions. The problem is only Case 4. We can discuss how to proceed once the first round is completed.

[nicolevasilevsky]

We tried to document some of the ideas of a disease vs phenotype here: https://mondo.monarchinitiative.org/pages/faq/#phenotype

(Talking to @matentzn now) This may not discriminatory enough though to help curators clearly decide if a term is a disease or phenotype though.

This is important for our collaborators at Open Target (cc @zoependlington) and my team at C-Path (cc @Daniel-Olson)

[pnrobinson]

There are things that are between HPO and Mondo and might require a differnet kind of integration, but this is a great start, thanks!

[matentzn]

@pnrobinson the next step is to provide a clear documentation of disease vs. phenotype. Do you have anything written / can provide something in written form? The text should enable a scientific curator to be able to recognise if a concept mentioned in an article is supposed to refer to the one, or the other.

[pnrobinson]

I will add something to the new HPO documentation. I wrote something like this several times and it keeps getting lost :-0.

[matentzn]

Awesome, thank you!

[matentzn]

@nicolevasilevsky @Daniel-Olson

Next steps:

• Read and comment on https://docs.google.com/document/d/1TEm91wDMrnygCpm-dVlJ-F9HJL-lfxwJveZZJSc5hFA/edit
• Develop a few actionable criteria from that text you can use to make a proper distinction
• Curate this set of 20 mappings I have selected for you and see if you can figure out which "case" they are: https://docs.google.com/spreadsheets/d/1QtefFH-ZahYe_8_dP4uvXbbw-DOxu1A0JAaMUAnUBVY/edit#gid=772835898
• In our next meeting we talk about what you found!

[pnrobinson]

Another thing that is still wrong 5 years after we have discussed this is the disease vs disease locus issue.

pancreatic cancer, susceptibility to, 2

is not a separate disease, it refers to a risk allele for pancreatic cancer. There are many other examples. It would be good to invite @ahamosh to the discussion since it is a little difficult to automatically determine whether a "susceptibility" is strong enough to be inherited as a Mendelian trait that can be observed in pedigrees and has additional recognizable profiles (e.g. BRCA gene are not only risk factors for breast cancer but have distinct profiles for ovarian), or whether, as in this case, I believe, the locus is essentially a strong GWAS hit but should not be regarded as a separate clinical disease.

[nicolevasilevsky]

@pnrobinson This is in the 'inherited disease susceptibility' branch but is also inferred to be a child of 'hereditary neoplastic syndrome', which is defined as:

The inherited predisposition toward getting a tumor.

Logical definition:

'inherited disease susceptibility'
 and ('predisposes towards' some neoplasm)

[pnrobinson]

We should prob discuss with Ada. The disease is the actual tumor and the predisposition is only visible in families. I do not think we are currently consistent about this

[pnrobinson]

Also, I had a great convo with Nico today and understand now what the unclear points where -- I will add some more documentation with examples this week