Re: Disease Phenotypes
Chris Mungall
I've made a couple of GUI changes, and some database changes
- I've created more named class expressions, so you should now see
nested class expressions for things such as "degenerate population of
dopaminergic cells"
- I've add a few rules such as one that looks for patterns
<org> bears <ph> inheres_in <pop> has_part <cc>
<q> inheres_in <pop>
and then creates class expressions linked to from the organism
<ph> instance_of [ <qType> that inheres_in <popType> that has_part
<ccType> ]
This brings the modeling more into line with what we've done for the
OMIM annotations. This deserves a bit more discussion, probably face
to face with a whiteboard
There are still a few more patterns to implement
The UI is still hugely opaque but the creation of these named
expression hopefully makes it slightly more intuitive. Oh, you can see
your comments attached to the instances -- that was a massive
oversight before
Similarity queries:
Let's look at human2b
http://www.berkeleybop.org/OBDUI/obdbirn/html/node/BIRN_PDPO%3Abirnlex_516_4
Click 'find similar':
http://www.berkeleybop.org/OBDUI/obdbirn/html/node/BIRN_PDPO%3Abirnlex_516_4/blast
Click the score on the 2nd one down:
human2b and human32 are similar by virtue of both bearing degenerate
populations of SN dopamine cells. This is relatively rare, so it
receives a decent score.
You can see all animals that bear this quality here:
(expand 'annotations')
3 humans and one mouse strain.
I'll try and find some examples that use multiple phenotypes per
organism, that's all for today though.
Chris Mungall
On Jan 2, 2009, at 11:01 AM, Sarah Maynard wrote:
> Hi Chris,
>
> I hope you had a great holiday and I am looking forward to seeing
> you at the end of the month.
>
> I am looking over the phenotypes on OBD again and have some questions:
>
> I see how I can find similar nodes by looking at humans, but when
> I click on a phenotype and click find similar nodes, no results
> appear. I would have thought that Phenotype2b and Phenotype32 would
> be similar, since on the organism level they are similar. Or a
> better example, if I navigate to Substantia Nigra Dopaminergic Cell
> 002b (http://www.berkeleybop.org/OBDUI/obdbirn/html/node/BIRN_PDPO%3Asao34994437_6
> ) or Population 002b (http://www.berkeleybop.org/OBDUI/obdbirn/html/node/BIRN_PDPO%3Asao2254405550_5
> ) and click find similar nodes, I thought I would get results of
> other Substantia Nigra Cells that degenerate as well.
Hi Sarah
At the moment the similarity search works against specially designated
relations. So you can compare two individual organisms with respect to
their phenotypes, but not two phenotypes. This is reasonably
configurable though.
> I am a little confused how you got to this link (http://www.berkeleybop.org/OBDUI/obdbirn/html/node/BIRN_PDPO%3APhenotype%5EOBO_REL%3Ainheres_in(sao%3Asao2254405550%5EOBO_REL%3Ahas_part(sao%3Asao34994437))
> )
This page is for the class expression representing "phenotypes of
populations of SN Dopa cells". The class expression was materialized
on the basis of instance data
> Also, is the quality all the phenotypes that have to do with
> Substantia Nigra Dopaminergic cell or SN Dopa Cell that
> degenerates. It isn't clear from the header.
The former, although this subsumes phenotypes that specifically
involve degenerated SN Dopa cells (as exemplified by human 002b)
> Also, how was the maxIC calculated? Is it normalized against
> anything?
I'll send a copy of an early draft of a paper later on.
For now, there is a description in the javadocs, though this is a more
API-level description:
http://www.berkeleybop.org/obd/docs/org/obd/model/stats/SimilarityPair.html
>
> One more thing: I didn't want to forget to understand more about
> how you calculated the similarity between phenotypes. From what you
> are saying it sounds like the similarity comparison takes into
> account how many other similar phenotypes we have. So, if there are
> two phenotypes that are similar, they would have a lower "score"
> than say 5 phenotypes being very similar. Is this right?
This is correct. The various metrics differ in how they deal with
multiple phenotypes. For the basic similarity score, having different
phenotypes will result in a penalty. However, for the maxIC, only the
two best matching phenotypes from two individuals are considered.
Again, more in the paper.
Cheers
Chris
>
> Thank you so much,
> Sarah