[Go] Change for GOA UniProtKB GAF file./ A suggestion

[Valerie Wood]

Slightly related, what is the long term strategy for getting IEA data 
into AmiGO?
A the main problem is the volume of annotations I have a suggestion:

For pombe we only include the IEA mappings in the data set provided to 
GO when they are non redundant with existing annotations.
In 2006 there were ~30000 electronic mappings, and ~15000 were retained
Today there are 55939 mappings and 4686 are retained.

For example tim44 has the following mappings:
 From IPR007379
Process       GO:0006886 intracellular protein transport
Function     GO:0015450 P-P-bond-hydrolysis-driven protein transmembrane 
transporter activity
Component     GO:0005744 mitochondrial inner membrane presequence 
translocase complex
 From IPR005682
GO:0006886 intracellular protein transport
Function     GO:0015450 P-P-bond-hydrolysis-driven protein transmembrane 
transporter activity
Component     GO:0005744 mitochondrial inner membrane presequence 
translocase complex
 From SP-KW
intracellular protein transmembrane transport
ATP binding

Only the mapping to ATP binding is retained as all of the others are 
covered by the manual annotation

Other genes have many more redundent mappings, for example top2 has 60 
mappings including 7 Interpro domains mapping to the same GO:0003677. 
These 60 mappings are fully represented by the 12 manual experimental 
annotation.

This procedure has a number of advantages

i) Clearer for Users
 It removes a massive over-presentation of data to the user.
I cannot see any major advantage in presenting  redundant mappings to 
the user.

ii) Quality control.
Because the curator is not presented with so many mappings, and complete 
annotation should, in theory, cover  the mappings (except in a minority 
of cased, it should be  possible to make an ISS to a characterised 
ortholog).
By following this annotation protocol, spurious mappings are easily 
identified and can be filtered and fixed.
Many 100's of mappings have been fixed in this way
http://sourceforge.net/tracker/?atid=605890&group_id=36855&func=browse
This also alters to problems in the ontology files,  if a parent is 
accidently removed, and this parent contains a valid mapping, the 
annotation will 'reappear', alterting the curator to problems with the 
ontology (this doesn't happen very often but it does provide an addition 
layer of QC)

ii) Space
It would generally reduce the size of the mapping file.
I have no idea of the size reduction. The reduction for pombe is > 90% 
but this is because the annotation coverage is high.
However, even un-annotated organisms could have an associated reduction 
in mappings, if only the most granular mapping is retained.
The number of IEAs will increase,  as you can see above the pombe 
mappings have doubled in the past couple of yeaars, but most mappings do 
not add any new information to the annotation.

Just a suggestion,
Val


Mike Cherry wrote:

> This afternoon the software group agreed to changing how we store the  
> goa_uniprot GAF file.  The large file will still be removed from 
> CVS.   This is okay because the file is too big for CVS and cannot 
> currently  be retrieved.  This file will still be available via to GO 
> FTP and  from the EBI FTP.  Both the submitted and filtered 
> goa_uniprot files  will be removed from CVS.  A new filtered file will 
> be created that  has all the IEA annotations removed and this file 
> will be in the CVS  repository.  Suggestions for this new file's name 
> are welcome, we were  thinking of : gene_association.goa_uniprot_noiea.gz
>
> Removing the files from CVS will happen almost immediately as  
> mentioned above you cannot get it from CVS anyway.  The older version  
> of the goa_uniprot file are available from the EBI FTP site.  The  
> files will still be available via HTTP and FTP at  
> www.geneontology.org.  This change simply means they will not be  
> obtainable via a checkout from CVS.  I'll work on creating the new  
> noiea file and add it to CVS next week.
>
> -Mike
>
> _______________________________________________
> Go mailing list
> Go at geneontology.org
> http://fafner.stanford.edu/mailman/listinfo/go
>



-- 
 The Wellcome Trust Sanger Institute is operated by Genome Research 
 Limited, a charity registered in England with number 1021457 and a 
 company registered in England with number 2742969, whose registered 
 office is 215 Euston Road, London, NW1 2BE.