[annotation] [Fwd:What evidence code to use?]
Judith Blake
jblake at informatics.jax.org
Thu Nov 29 09:10:12 PST 2007
I shouldn't have jumped into this. But....
ISS for MGI requires that the ISS be backed up with experimental data.
Clearly, the analysis brought forward does not do that.
RCA from SGD perspective requires experimental data sets. From MGI
perspective, was used for the FANTOM analysis (only) when the sequence
analysis was part of expert annotation. MGI has not had much occasion
to use RCA since the Fantom, and we are gradually removing these.
The argument about ISS was whether it was to be restricted to use with
orthologs that had experiments or whether it was to include sequence
analysis and HMM type studies done in the individual organisms.
We resolved that, I thought, by moving toward ISS with subcodes of ISO
(for orthology sets) and IS- (I don't remember) for HMMs and other
supervised sequence analysis. The study brought forward by Tanya could
be either the ISS (generic sequence analysis) or the other one, but
certainly these are not backed by experimental data, so with the current
RCA, these could best, perhaps, be
ISS (generic) but we don't have this implemented yet
IEA.....why not? well, it's not just an electronic analysis...
Again, these reflects only predictive analysis, there is no experimental
data, MGI would prefer ISS only be used when backed by experimental data
(or the new category) and SGD would prefer that RCA be restricted to
experiment +/- computational analysis using sequence.
In the end, I would like to express my thoughts again that we should not
drown ourselves in this discussion. By going to the reference or by
reading MOD supplied abstract, users can determine the predictive
algorithm source if they want too. One could argue that we spend too
too much time on sorting this out when we do have group consensus that
evidence codes are mostly to provide clues to users as to the assay
generic classes that the annotation is supported by. The reference is
really the source, and we toe a fine line between just using
'experimental' and 'predicted', and providing all the gory details of
the analysis.
Cheers,
Judy
Pascale Gaudet wrote:
> But, I thought RCA required experimental data??
>
> From documentation: http://www.geneontology.org/GO.evidence.shtml#ica
>
> * Predictions based on computational analyses of large-scale
> experimental data sets
> * Predictions based on computational analyses that integrate
> datasets of several types, including experimental data (e.g.
> expression data, protein-protein interaction data, genetic
> interaction data, etc.), sequence data (e.g. promoter sequence,
> sequence-based structural predictions, etc.), or mathematical
> models
>
> Pascale
>
> Judith Blake wrote:
>> ok with me if we need to make the distinction. I took it to mean the
>> difference between a simple alignment report and a more
>> comprehensive analysis. Phylogenetic analyses employ powerful
>> algorithms, but at the core of the analysis are manually curated
>> multiple alignments from hundreds of species. These could be RCA for
>> me. At the end of the day, I think it doesn't matter :) since all
>> these measures are predictive and not experimental determinations.
>>
>> Judy
>>
>>
>> Karen Christie wrote:
>>> My recollection is that RCA was proposed by SGD to handle papers
>>> such as Samanta and Liang 2003 (url below) where they did
>>> computational analysis of large-scale protein interaction data.
>>>
>>> http://db.yeastgenome.org/cgi-bin/reference/reference.pl?dbid=S000074191
>>>
>>>
>>> The original documentation for RCA explicitly stated that it was not
>>> to be used for sequence data. At the St. Croix meeting, Sue Rhee
>>> brought up the point that some computational analyses combined
>>> sequence data into the types of analyses done by Samanta and Liang.
>>> On that basis, it was agreed that RCA could include sequence data,
>>> but was not intended for analyses that were entirely sequence based.
>>>
>>> -Karen
>>>
>>>
>>> On Wed, 28 Nov 2007, Mike Cherry wrote:
>>>
>>>> I believe RCA was proposed by SGD to use with analyzes like Biopixie.
>>>>
>>>> Cheers, Mike
>>>>
>>>>
>>>> On Nov 27, 2007, at 9:00 PM, Judith Blake
>>>> <jblake at informatics.jax.org> wrote:
>>>>
>>>>> This is exactly what RCA was originally used for. With the FANTOM
>>>>> project [mouse full length cDNA annotatons], participants employed
>>>>> a series of algorithmic approaches combined with manual inspection
>>>>> and evaluation to provide annotations. Actually, I think RCA was
>>>>> created as a result of the FANTOM project.
>>>>>
>>>>> Judy
>>>>>
>>>>> Tanya Berardini wrote:
>>>>>> Forwarding this from the evidence code discussion group.
>>>>>> Apologies to those who are on both lists. I've sorted the emails
>>>>>> from top to bottom in chronological order for easier reading:
>>>>>>
>>>>>> ----------
>>>>>> My original email:
>>>>>>
>>>>>>> Ah, the eternal question: Is it ISS, is it RCA?
>>>>>>>
>>>>>>> I've got a paper that describes the identification of a nice big
>>>>>>> set
>>>>>>> of transcription factors in Arabidopsis.
>>>>>>>
>>>>>>> http://www.ncbi.nlm.nih.gov/sites/entrez?cmd=Retrieve&db=PubMed&list_uids=11118137&dopt=AbstractPlus
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> The authors use a combination of motif searches + BLAST + sequence
>>>>>>> alignment and review those by eye and came up with 1500 or so genes
>>>>>>> that they call 'transcription factors.'
>>>>>>>
>>>>>>> Right now, we've got these annotated to 'transcription factor
>>>>>>> activity' with the evidence code ISS but nothing in the
>>>>>>> evidence_with
>>>>>>> column. If I leave these as ISS, I'd like to put something in the
>>>>>>> with column, but what? Does this type of a combination of sequence
>>>>>>> analysis methods that's reviewed manually make it RCA? Not
>>>>>>> according
>>>>>>> to the current RCA documentation:
>>>>>>>
>>>>>>> "Examples where the RCA evidence code should not be used:
>>>>>>>
>>>>>>> * Annotations based on more than one type of gene product
>>>>>>> sequence
>>>>>>> based evidence, including such things as BLAST, profile HMMs,
>>>>>>> TMHMM,
>>>>>>> SignalP, PROSITE, InterPro, mapping files such as interpro2go etc.
>>>>>>> should use the ISS code. "
>>>>>>>
>>>>>>> Should I wait till ISS comes to a resolution?
>>>>>>>
>>>>>>> Help!
>>>>>>
>>>>>> ---------
>>>>>> Ben's reply:
>>>>>>
>>>>>> If you can't put something USEFUL in the WITH column, I think
>>>>>> this has to be RCA.
>>>>>> I guess under the new, non-documented system, this would be
>>>>>> ISS/no "With" ISA/ISO/ISM would require withs... (either seq ids
>>>>>> or model aka interpro ids).
>>>>>>
>>>>>>
>>>>>> Ben
>>>>>>
>>>>>> ----------
>>>>>>
>>>>>> Val's reply:
>>>>>>
>>>>>> This is *exactly* the type of data why I was orginally suggesting
>>>>>> that RCA should not be restricted to analysis which include some
>>>>>> experimental component. Unfortunately I couldn't come up with
>>>>>> any good examples at the time.
>>>>>>
>>>>>> These would surely be better as RCA, even though they are
>>>>>> sequence based
>>>>>>
>>>>>> Val
>>>>>>
>>>>>> ----------
>>>>>>
>>>>>> Susan's reply:
>>>>>>
>>>>>> I've just hit another example...
>>>>>>
>>>>>> Enhanced function annotations for Drosophila serine proteases: A
>>>>>> case
>>>>>> study for
>>>>>> systematic annotation of multi-member gene families.
>>>>>>
>>>>>> Shah PK, Tripathi LP, Jensen LJ, Gahnim M, Mason C, Furlong EE,
>>>>>> Rodrigues V,
>>>>>> White KP, Bork P, Sowdhamini R.
>>>>>>
>>>>>> PMID: 17996400
>>>>>>
>>>>>> This is a functional classification of serine proteases based on a
>>>>>> 'function residue clustering' algorithm. The algorithm
>>>>>> incorporates info
>>>>>> from sequence alignments, hydrophobicity plots and info about key
>>>>>> residues from 3D structures - all sequence based but no one thing
>>>>>> to put
>>>>>> in the 'with'.
>>>>>>
>>>>>> Susan
>>>>>>
>>>>>> -----------
>>>>>>
>>>>>> Pascale's reply:
>>>>>>
>>>>>> Tanya,
>>>>>>
>>>>>> I thought we agreed that BLAST and InterPro were ISS, as you
>>>>>> point out. I don't think ISS + ISS = RCA?? That is, I would say
>>>>>> using InterPro or the BLAST result should be enough to make the
>>>>>> annotation; we dont need to capture both? In this case, the
>>>>>> easiest might be using ISS with an InterPro domain ID in the 'with',
>>>>>>
>>>>>> Similarly in the paper Susan cites, they mention several domains
>>>>>> and also they have compared to several proteins whose 3D
>>>>>> structure has been determined hence can be used in the 'with' - I
>>>>>> would pick one of those example proteins and ISS to that.
>>>>>>
>>>>>> Pascale
>>>>>>
>>>>>> ---------
>>>>>>
>>>>>> Any other thoughts?
>>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Tanya
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> -------- Original Message --------
>>>>>> Subject: Re: [evidence] What evidence code to use?
>>>>>> Date: Wed, 21 Nov 2007 08:43:16 -0500
>>>>>> From: Pascale Gaudet <pgaudet at northwestern.edu>
>>>>>> Reply-To: pgaudet at northwestern.edu
>>>>>> Organization: Northwestern University
>>>>>> To: tberardi at acoma.stanford.edu
>>>>>> CC: evidence at genome.stanford.edu
>>>>>> References: <47437C88.5070204 at acoma.stanford.edu>
>>>>>>
>>>>>> Tanya,
>>>>>>
>>>>>> I thought we agreed that BLAST and InterPro were ISS, as you
>>>>>> point out.
>>>>>> I don't think ISS + ISS = RCA?? That is, I would say using
>>>>>> InterPro or
>>>>>> the BLAST result should be enough to make the annotation; we dont
>>>>>> need
>>>>>> to capture both? In this case, the easiest might be using ISS
>>>>>> with an
>>>>>> InterPro domain ID in the 'with',
>>>>>>
>>>>>> Similarly in the paper Susan cites, they mention several domains and
>>>>>> also they have compared to several proteins whose 3D structure
>>>>>> has been
>>>>>> determined hence can be used in the 'with' - I would pick one of
>>>>>> those
>>>>>> example proteins and ISS to that.
>>>>>>
>>>>>> Pascale
>>>>>>
>>>>>>
>>>>>>> ------------------------------------------------------------------------------------------
>>>>>>>
>>>>>>> Tanya Berardini, Ph.D. tberardi at acoma.stanford.edu
>>>>>>> The Arabidopsis Information Resource FAX: (650) 325-6857
>>>>>>> Carnegie Institution of Washington Tel: (650) 325-1521 ext. 325
>>>>>>> Department of Plant Biology URL: http://arabidopsis.org/
>>>>>>> 260 Panama St.
>>>>>>> Stanford, CA 94305
>>>>>>> ------------------------------------------------------------------------------------------
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>
>>
>>
>>
>
> --
> ~~~~~~~~~~~~~~~~~~~
> Pascale Gaudet, PhD
> Scientific Curator, dictyBase
> Northwestern University, Chicago, IL
> pgaudet at northwestern.edu
> www.dictybase.org
> ~~~~~~~~~~~~~~~~~~
More information about the Annotation
mailing list