Handling apparent contradicitions with negation data

[cmungall]

One strategy for generating negation data is to derived disjointness axioms based on PATO definitions. E.g. PATO already has:

'increased size' = size and inc-relative-to SOME normal 'decreased size' = size and dec-relative-to SOME normal

So we can heuristically generate

'increased size' and inh some femur DISJOINT-WITH  'decreased size' and inh some femur

However, our approach will need to be more subtle due to the incoherencies this generates: we need instead to have a probabilistic DisjointWith

UNSAT: HP:0007494 'Discrete 2 to 5-mm hyper- and hypopigmented macules'
UNSAT: HP:0007471 'Axillary and groin hyperpigmentation and hypopigmentation'
UNSAT: HP:0009123 'Mixed hypo- and hyperpigmentation of the skin'
UNSAT: HP:0007441 'Hyperpigmented/hypopigmented macules'
UNSAT: HP:0007450 'Increased groin pigmentation with raindrop depigmentation'
UNSAT: HP:0007402 'Areas of hypopigmentation and hyperpigmentation that do not follow Blaschko lines'
UNSAT: HP:0007509 'Patchy hypo- and hyperpigmentation'

[cmungall]

Also consider: Triploidy

wide mouth: Hallmark 90% narrow mouth: occasional 7.5%

This is logically consistent but a naive translation to OWL will lead to inconsistency

[pnrobinson]

Note that it is indeed possible for a patient to have both hyper and hypo pigmentation, and inference procedures should not assume that these terms refer to the entirety of the skin (in which case it would be a contradiction).

[jmcmurry]

Is there more to supporting negation than is mentioned above? From R24 "We will develop algorithms that will weight feature frequencies and negative annotations using our Bayesian ontology-querying framework89, which we will extend from a purely HPO-based algorithm to a cross-species inference algorithm accounting for species differences. Our probabilistic framework also provides a natural framework for integrating quantitative data; existing phenotype ontologies provide ready-made ‘bins’ for quantitative phenotypes, such as ‘increased body weight’ or ‘abnormal femur size’. Terms such as these are used both in the model organism literature, and by the curators of model organism databases. Sometimes this is a judgment call on the part of a researcher, but in the case of many phenotyping pipelines, normal and abnormal ranges of measurements are defined as part of the protocol. We will use these protocols and the data gather from phenotyping pipelines to learn and interpolate normal ranges for a variety of phenotypes. This will allow us to combine both qualitatively assigned abnormal ranges with quantitative data. We will also improve our capacity for differential diagnostics as well as for novel disease gene discovery by including these new advanced scoring metrics within our website as well as in our tools such as Exomiser."

[pnrobinson]

@drseb @jmcmurry @cmungall @mellybelly There is some overlap to the goals of my Hipbi application that I think in spirit should be part of Monarch. I think that even by only adding negative annotations, BOQA's performance should increase. Seb had some interesting ideas for BOQA that we discussed in Berlin. Perhaps he can comment on whether any of this is of interest for him. I have some ideas for this that are probably pretty obvious but we should make a roadplan for this.

[jmcmurry]

Great, thanks Peter. If there are any Hipbi tickets can you crosslink please?

[drseb]

We basically have a first crappy version of this in the opposite-of paper. Also somebody from Mikes group already did it.

[pnrobinson]

I wouldn't say crappy, but what exactly does Mike's group have?

[cmungall]

We can count this ticket as done with the opposites-of paper.

https://www.biorxiv.org/content/early/2017/07/27/108977

[jmcmurry]

To really count this done we need to be able to say we've incorporated negation into the matching algorithms (Phenol and Ontobio) and have leveraged it in annotation pipelines.

[jmcmurry]

cc: @monarch-initiative/algorithms team to tackle this.

[pnrobinson]

Phenol handles negation but this is not an algorithm. Other slgos can run on phenol though

Sent from my iPhone

On Aug 2, 2018, at 11:24 AM, Julie McMurry notifications@github.com<mailto:notifications@github.com> wrote:

To really count this done we need to be able to say we've incorporated negation into the matching algorithms (Phenol and Ontobio) and have leveraged it in annotation pipelines.

- You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/monarch-initiative/owlsim-v3/issues/13#issuecomment-409965305, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AEtuPI2TDrALvAeqcGOOEvKJY49mBliVks5uMxmGgaJpZM4EOYom.

The information in this email, including attachments, may be confidential and is intended solely for the addressee(s). If you believe you received this email by mistake, please notify the sender by return email as soon as possible.