[MOBY-l] ideas from the CRIB

[Heiko Schoof]

Hi Mark,
happy to see my ideas mirrored.

To get some grip on the vastness of relationship space, I was 
immediately talking about an ontology... meaning, we can start with 
basic classes, and anyone who wants to get finegrained can go down the 
hierarchy forever if he wants, as long as toplevel users can still use 
the basic classes and get what they expect.

To design an ontology, though, is a mindblaster of a problem, especially 
if it's to stand up to trial without having to be changed. For sure it'd 
need to make sure it doesn't integrate orthogonal ideas. I'm afraid that 
your "Primary" and "Secondary" reference thing already rings all alarm 
bells with me there: The concept of whether a relationship is firsthand 
or more cirmumstantial is, in my view, orthogonal (i.e. a different 
means of classification) from the type of relationship, e.g. isa, hasa. 
And should thus go into a separate ontology. To your concrete example:

MyLiverCancerGene isReferredToIn PubMedCitation1
MyLiverCancerGene isA TyrosineKinase
MyLiverCancerGene has KeywordLiverCancer
TyrosineKinase isReferredToIn PubMedCitation2
LiverCancer isReferredToIn PubMedCitation3

are all valid statements in my view. What I understand you are 
suggesting is to put in a CRIB:

isReferredToIn (primaryReference) PubMedCitation1
isA (primaryReference) TyrosineKinase
has (primaryReference) KeywordLiverCancer
isReferredToIn (secondaryReference) PubMedCitation2
isReferredToIn (secondaryReference) PubMedCitation3

Which I would strongly want to avoid, and I have already written the 
example in a way to show that the primary/secondary business is to my 
view separate from the relationship business. Instead, I'd put in the 
CRIB:

isReferredToIn PubMedCitation1
isA TyrosineKinase
has KeywordLiverCancer

And use MOBY to jump on from there, retrieving the TyrosineKinase object 
that contains in the CRIB:
isReferredToIn PubMedCitation2

and the KeywordLiverCancer object that contains:
isReferredToIn PubMedCitation3

What do we have Moby for, if not for following up those indirect 
relationships?

For the relationship types, we need to sit down with an information 
theorist, I'd guess. I keep starting with basic things like isA, partOf, 
hasA, but I never know when I'm complete.

isA: This contains all the identity, synonyms, heredity stuff. In Mark's 
terms: IdenticalTo, RelatedTo, FamiliarWith (except I don't agree about 
the individual exons, that's partOf. FamiliarWith in my reading means 
that I can move from one to the other object through some simple 
transformation, e.g. evolution, translation, and so on, but it's a 
one-on-one relationship (that doesn't exclude it occuring many times on 
one object))

partOf: where one is more than the other, one is contained in the other, 
e.g. exons and introns are parts of a transcript, domains are parts of a 
protein, loci are parts of a chromosome.

hasA: attachment, but not isA or partOf. This is all the annotation 
stuff, the references, keywords, GO terms, and so on.

I'd be already very happy if we implemented just that very basic 
distinction, three base classes of relationships, with the option to 
refine them later. It'd immediately allow to differentiate between 
annotation and identity. Maybe a distinction in the isA part is 
valuable, identicalTo as an isA that is in the same scope and level 
(e.g. an EMBL acc and a SWISSPROT ID) and inheritsFrom as an isA on a 
different level (e.g. an enzyme that inheritsFrom an EC class or an 
organism that inheritsFrom a species that inheritsFrom a genus or a 
protein "TYRK" that inheritsFrom a gene "tyrk" that inheritsFrom a 
genefamily "Tyrosine Kinases" that inheritsFrom "Kinases" etc...)

So, Mark, I even underbid your dozen types ;-)

Best, Heiko

------------------------------------
Dr. Heiko Schoof
Technische Universitaet Muenchen
-Genome Oriented Bioinformatics-
Wissenschaftszentrum Weihenstephan
85350 Freising
Germany

Tel. +49 8161 71 5632
Fax +49 8161 71 5629
h.schoof at wzw.tum.de
http://binfo.bio.wzw.tum.de