Triptych enhancements

[jmcmurry]

see https://github.com/monarch-initiative/monarch-app/issues/1417

[cmcnamara87]

Data structure: array of: 2 phenotypes

• similarity

[{
  phenotype1: { label: asdafds, informationContent: 7 }
  phenotype2: { label: 64hr, informationContent: 3 }
  similarity: 95
}, ... ]

[cmcnamara87]

from monarch-initiative/monarch-app#1417 @davidyu85

Have a look at b), with with a monochromatic scale, instead of multi-coloured.

[cmcnamara87]

I was thinking of maybe expanding the simscore to show the distance? Where you could hover over each hop, and see what it is.

[cmcnamara87]

I have no idea of stuff would match, that would contain like 100 hops :P But can try it out and see if its easier to understand, than a number

[cmcnamara87]

[jmcmurry]

Number of hops is usually 0 to 2. Beyond that I'm not sure it often scores as a match at all...

Sent from my iPhone

On May 3, 2017, at 21:03, Craig McNamara notifications@github.com wrote:

— You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[cmungall]

number of hops may be less informative than you think. In highly latticey parts of the ontology you can imagine a MRCA with high IC and many subnodes

[cmcnamara87]

@cmungall interesting. 4 high IC hops is really closer than 4 low IC hops. makes sense. ill see what we can do.

[davidyu85]

First draft version is now up and online:

https://genome-one.github.io/monarch/triptych/

[jmcmurry]

[jmcmurry]

Are there going to be in-common terms shown on hover?

[jmcmurry]

Please consider a systems-level overview (preferably similar to the triptych overview) Disease A and B in columns. Something like this:

wherein a sim score is calculated at the level of each system, in addition to the overall score.

As I envision it, clicking on any one system (anywhere in the row) would bring up the deeper triptych of matching of just that particular set of phenotypes in a category. The tricky thing is that there will be two very different use cases: 1) syndromes with little bits of phenotypes scattered across categories 2) more traditional diseases with more concentration in one or two systems. It is also important to realize that some phenotypes have multiple parentage in different systems, eg. some eye phenotypes are both neurological and physiological. We should think about how to represent those?

For both the system-level view and the leaf-node view, it would be great to have the ability to list from best match to worst.

Ideally, it would also be good to sort from most important to least important, independent of whether match occurs. This gets at IC score but I wouldn't phrase it this way. Also in time we may want to tweak according to population frequency etc.

[davidyu85]

@jmcmurry I will be working on this table starting from today @cmungall With regards to api calls, as I am still new to this project and the initiative itself, can you please advise what api call I would need to get the data for this visualisation?

[jmcmurry]

Bowing of the long bones shows up in both Achondroplasia and Pseudoachondroplasia, but in both cases is unmatched. I haven't looked at source to see what the problem is.

[jmcmurry]

As we discussed at last huddle, it would be great if triptych made better use of vertical space. This also got me thinking also about the order in which phenotypes are listed. Space and order are separate concerns but which I've tried to address in same suggestion below. Let me know your thoughts. (Don't pay attention to the IC scores here; half are made up.) If we list most similar terms to most divergent, then non matches that are nevertheless related float to surface. This kind of view could be useful both for comparing patients and for doing differential diagnoses. There are lots of things that don't rise to the current cut-off for a match, but that are nevertheless interesting for instance genu varum and genu valgum below.

[mellybelly]

I like this a lot Julie. I wonder if some better categorization of the phenotypes would be helpful? I wonder if the bars could somehow show degree of similarity by their overlap, negating the need for the circles? Also we'll still want hover over to see the LCS, right?

[jmcmurry]

I wonder if some better categorization of the phenotypes would be helpful?

Agreed; the dynamic clustering already does this, however, it is hard to appreciate fully because of how hard it is to see more than a few terms at once.

[jmcmurry]

@cmcnamara87 https://genome-one.github.io/monarch/triptych/ is now not online anymore; any update?

[genome-one]

Hmm ... we haven't done anything with it ... We'll look into it.

[jmcmurry]

Any update?

[jmcmurry]

Additional feature request: when hovering over Genu Varum on right, also somehow emphasize Genu Valgum on left. Same for all matches that are significant but not the very best one.

[jmcmurry]

Here's my mockup from Peter's idea about having a more clinician-friendly cochrane-style view of phenotype profile match. Feedback welcome.

Currently, I have put the scale from does not support diagnosis, to supports diagnosis. However, there may eventually be the possibility of contraindicating diagnosis in cases where a patient has an asserted negation of a phenotype (particularly a hallmark phenotype) that is associated with a disease.

[jmcmurry]

Note that in the Cochrane, it is risk ratios where smaller is more significant. We could adopt that convention of left = significant if we prefer, even though we are showing something altogether different.

[genome-one]

@jmcmurry @cmungall Trying to get the data for the image above - but can't seem to find an appropriate call in the BioLink API for the profile to profile comparison. Any help would be greatly appreciated.