PubChem Relational Properties and Relationship Hierarchy

[Gang Fu]

Hi,

Please fine the attachment about the definitions and descriptions on PubChem relational properties and relationship hierarchy. The basic semantics of relational properties were based on the guidelines prepared earlier by Janna and Evan. Any suggestion to improve the relationship hierarchy is highly appreciated.

Thank you very much!

Best,

Gang

[Michel Dumontier]

Janna,

 There seems to be some overlap with the proposal with ChEBI relations.  How do you want ChEBI to play into this?

m.

--

Michel Dumontier

Associate Professor of Bioinformatics, Carleton University

Chair, W3C Semantic Web for Health Care and the Life Sciences Interest Group

http://dumontierlab.com

[Janna Hastings]

The overlap with ChEBI relations is OK. Where there is an overlap,
the same relationship name should be used, and we will make it
explicit that the ChEBI and PubChem relations have the same semantics.

I've inserted a few comments in the attached version of the document
circulated by Gang.

Cheers, Janna

[Egon Willighagen]

Open Phacts use VoID + PROV tot define kink sets... When back home I will try to remember to send details... But you can try Google too, "open phacts Alasdair void linkset"

Egon

Op 17 dec. 2012 17:31 schreef "Gang Fu" <gangf...@gmail.com> het volgende:

[Egon Willighagen]

On Mon, Jan 7, 2013 at 6:37 PM, Egon Willighagen
<egon.wil...@gmail.com> wrote:
> Open Phacts use VoID + PROV tot define kink sets... When back home I will
> try to remember to send details...

http://www.openphacts.org/specs/2012/WD-datadesc-20121019/

This specification is very much oriented at linking things that are
the same *at some level*... e.g. tautomers, salt-parent relations,
etc...

Egon


--
Dr E.L. Willighagen
Postdoctoral Researcher
Department of Bioinformatics - BiGCaT
Maastricht University (http://www.bigcat.unimaas.nl/)
Homepage: http://egonw.github.com/
LinkedIn: http://se.linkedin.com/in/egonw
Blog: http://chem-bla-ics.blogspot.com/
PubList: http://www.citeulike.org/user/egonw/tag/papers

[Gang Fu]

Hi,

I added some corrections and explanations (please find attachment). It should be unambiguous to define "has-standardized-form", "has-parent", "has-component" relationships, although the name of these predicates can be changed :)
It will be a real challenge to define "has-similar-structure" relationship, which is quite variable. I understand it has many way to define structural similarity, according to different descriptors, index, and metrics, etc..
I would suggest the semantics for structural similarity is database-specific. I suggest we could have a generic term such as "has-similar-structure", underneath it, we could have database-specific sub-properties, such as "has-PubChem-2D-similar-structure", "has-PubChem-3D-similar-structure", and so on. This proposal derives from the fact that similarity information provided by database provider depends on their own definition and are quite different from each other, right?:) User can define their own structural similarity, but will not be conflict with database provider information, since it can be overwritten for user-specific purpose.

We can discuss more about the relationship hierarchy, once the proposal is accepted by CHEMINF.

Best,
Gang 

[Gang Fu]

Hi Egon,

To me, this specification is for interlinked datasets, right?:). It would definitely be useful when we want to provide PubChem RDF and link it to other dataset RDF. But the compound-compound relationship I proposed here is not for cross-dataset resources, but for within PubChem dataset. It should be beneficial for ChEBI and other datasets, if a more general compound relational hierarchy can be well defined in CHEMINF.

Best,
Gang

[Michel Dumontier]

On Thu, Jan 10, 2013 at 8:16 AM, Gang Fu <gangf...@gmail.com> wrote:

"has-similar-structure" relationship, which is quite variable. I understand it has many way to define structural similarity, according to different descriptors, index, and metrics, etc..
I would suggest the semantics for structural similarity is database-specific.

In order to more accurately define the terms and conditions of the similarity, you basically need another object that will link the two molecules and also assert the amount of similarity and/or the methods used to determine it. This kind of semantic description is possible using SIO:

:a rdf:type :structural-similarity

 sio:refers-to pubchem_compound:x;

 sio:refers-to pubchem_compound:y;
 sio:has-attribute [ a :tanimoto-score; sio:has-value "0.9"];

 sio:has-source :database;

 sio:is-output-of [ a :software-execution-process; sio:has-agent :some-software; sio:has-parameter [a :parameter; sio:has-value value xx]]

...

or whatever you want to associate with this object ...

m.

[Gang Fu]

Hi Michel,

This is very good suggestion! Thank you very much!

Best,
Gang