[BioC] GOstats: Listing genes from hyperGTest
Marc Carlson
mcarlson at fhcrc.org
Fri Oct 24 18:52:43 CEST 2008
Hi Jim,
Thanks again for swooping in to get these guys answer while those of us
on the west coast are still sleeping. We really do appreciate it. To
answer your question, the mapping org.Hs.egGO does not contain all
relationships between all GO terms and all genes. What it is supposed
to contain are the terminal mappings (ie. the most specific "leaves"
from the GO DAG) that can be mapped onto human genes. This is the
information as it was originally input by curators and stored at NCBI.
>From that we can derive that if a terminal (ie. more specific term) is
related to a gene, that its less specific parent terms should also be
mapped to it, and this information is stored in the org.Hs.egGO2ALLEGS
mapping. So as you pointed out you can find these results by doing:
get("GO:0032502", org.Hs.egGO2ALLEGS)
#and
get("GO:0032501", org.Hs.egGO2ALLEGS)
I can't tell you why nobody annotating genes in human ever specifies
these two terms, I suspect it is because being as general as they are
they are not very meaningful. So I am kind of glad that I don't see
them used, because it tells me that the annotations are more specific
than these two terms. And more specific annotations should be more
useful than vague generalized ones. I guess it is also possible that
NCBI could do some filtering on their term mappings before they share
them with the world, but I don't think so because they do (on rare
occasion) have genes from other organisms that are mapped to both of
these terms.
As for the complete collection of GO DAG terms, the entirety of that
information should be found only in the GO.db package since not every
piece of GO should necessarily be represented by a gene in humans that
is known to be involved in that process/component/function. And when I
check, I do see these two terms there so that part is also ok. So to
clarify, the GO information in org.Hs.eg.db is only meant to indicate
how specific genes map onto the GO ontology, which is very unlikely to
correspond to all the GO terms being represented, in contrast the
structure of GO and the complete listing of GO terms is found in GO.db.
If this is all still confusing, please let me know. The GO stuff is
confusing.
Marc
James W. MacDonald wrote:
> Hi Tim,
>
> Hmm. That's weird.
>
> First off, I think you want to be using the GO2ALLEGS mapping rather
> than the GO2EG (or revmap(org.Hs.egGO) which is equivalent).
>
> The reason for this is that the GO ontology is a directed acyclic
> graph where all children are subsets of their parents, so by
> definition all genes that are appended with a child GO term are also
> appended with the parent term (e.g., multicellular organismal
> development is a part of the developmental process, so any genes
> involved with the former are by default involved with the latter as
> well).
>
> So modulo differences in our BioC versions, I get the same results as
> you:
>
> > summary(hyp, categorySize=2000)
> GOBPID Pvalue OddsRatio ExpCount Count Size
> 1 GO:0032501 0.0001909763 3.000224 11.303450 23 3420
> 2 GO:0032502 0.0008306266 2.730994 9.984714 20 3021
> 3 GO:0007275 0.0035077569 2.576107 7.224954 15 2186
> 4 GO:0007154 0.0042870538 2.271198 13.058459 22 3951
> Term
> 1 multicellular organismal process
> 2 developmental process
> 3 multicellular organismal development
> 4 cell communication
>
> And can get the genes using org.Hs.egGO2ALLEGS:
>
> > geneIds[geneIds %in% get("GO:0032501", org.Hs.egGO2ALLEGS)]
> [1] "1281" "1410" "157506" "1906" "2625" "3043" "3553" "4015"
> [9] "4885" "4886" "5021" "5156" "5788" "6005" "6507" "673"
> [17] "6876" "695" "7026" "7070" "7412" "800" "83890"
>
>
> What I don't understand is how multicellular organismal process and
> developmental process can be missing from the org.Hs.egGO map
>
> Both these processes are direct children of biological process, so
> really they should be there. Maybe Marc Carlson can shed some light on
> this.
>
>
> Best,
>
> Jim
>
>
>
> Tim Smith wrote:
>> Thanks Jim. I'm still having problems, i.e., I cannot find which
>> subset of input genes resulted in the significant GO term. I have
>> reproduced the problem that I am having:
>>
>> -----------------------------------------------------
>> library(org.Hs.eg.db)
>> library(GOstats)
>> library(GO.db)
>>
>> # Set of genes (Entrez Ids) that I want to investigate
>> geneIds <- c("10406", "10418", "11082", "1281", "1410",
>> "157506", "167410", "1906" , "2029", "23091", "2625" , "2823" ,
>> "2877", "2993", "3039" , "3043", "3046", "3283", "3553",
>> "4015", "4069", "4258", "4345", "4353", "4885", "4886",
>> "5021", "5055", "5151", "5156", "5320", "5553", "55885",
>> "56667", "5788" , "5999", "6005", "629", "6507", "653145",
>> "6590", "673", "6876", "695", "7026", "7070", "7103",
>> "7412", "760", "7738", "800", "828", "83890", "945",
>> "963")
>> ### I have reproduced Jim's code for the test
>> univ <- Lkeys(org.Hs.egGO)
>> param <- new("GOHyperGParams", geneIds = geneIds,
>> universeGeneIds=univ, annotation="org.Hs.eg.db", ontology="BP")
>> hyp <- hyperGTest(param)
>> summary(hyp,categorySize=2000)
>> ----------------------------------------------------
>>
>> The result I get is :
>>
>> GOBPID Pvalue OddsRatio ExpCount Count
>> Size Term
>> 1 GO:0032501 0.0001568145 3.021968 11.976486 24 3438
>> multicellular organismal process
>> 2 GO:0032502 0.0012459810 2.626265 10.297409 20
>> 2956 developmental process
>> 3 GO:0007275 0.0051457709 2.457897 7.517527 15 2158
>> multicellular organismal development
>> 4 GO:0007154 0.0061081457 2.187407 13.418681 22
>> 3852 cell communication
>>
>> I now want to know which subset of genes resulted in 'GO:0032501'.
>> The error I get is:
>>
>>> geneIds[geneIds %in% get("GO:0032501", revmap(org.Hs.egGO))]
>> Error in .checkKeys(value, Rkeys(x), x at ifnotfound) : value for
>> "GO:0032501" not found
>>
>>> geneIds[geneIds %in% get("GO:0032502", revmap(org.Hs.egGO))]
>> Error in .checkKeys(value, Rkeys(x), x at ifnotfound) : value for
>> "GO:0032502" not found
>>
>> [[elided Yahoo spam]]
>>
>> thanks a lot.
>>
>> Tim
>>
>>
>>
>>
>>
>> Hi Tim,
>> Yeah, probeSetSummary() is probably not what you want, if you are not
>> starting with an Affy chip. There are some gymnastics required to map
>> things back to the original Affy chip that you won't need to do. In
>> addition, if you are not using a conditional hypergeometric analysis,
>> it should be pretty simple to get what you want without even needing
>> to parse things out of the GOHyperGResult object. An example:
>> ## fake up some data
>>> geneIds <- Lkeys(org.Hs.egGO)[sample(1:5000, 500)] univ <-
>>> Lkeys(org.Hs.egGO) param <- new("GOHyperGParams", geneIds = geneIds,
>> universeGeneIds=univ, annotation="org.Hs.eg.db", ontology="BP")
>>> hyp <- hyperGTest(param) summary(hyp, categorySize=10)
>> GOBPID Pvalue OddsRatio ExpCount Count Size Term 1
>> GO:0007338 0.002723500 29.25101 0.07808304 2 54 single
>> fertilization 2 GO:0009566 0.002925855 28.16374 0.08097501 2
>> 56 fertilization
>> So we have two terms of interest. Getting the Entrez Gene IDs from
>> the input set that map to these terms is easy:
>>> geneIds[geneIds %in% get("GO:0007338", revmap(org.Hs.egGO))]
>> [1] "100131137" "10007"
>> Now you might also want to know which 54 Entrez Gene IDs map to that
>> particular GO term. Since you are not conditioning, this includes
>> that particular GO term and all its offspring.
>>> offspring <- get("GO:0007338", GOBPOFFSPRING) egids <-
>>> unique(unlist(mget(c("GO:0007338", offspring),
>> revmap(org.Hs.egGO), ifnotfound=NA), use.names=FALSE))
>>> egids[!is.na(egids)]
>> [1] "1047" "4179" "4240" "4486" "4809"
>> "5016" [7] "6674" "7783" "7784" "7802"
>> "7993" "8747" [13] "8748" "8852" "9082"
>> "10007" "10361" "22917" [19] "26476" "53340"
>> "57055" "57829" "64100" "93185" [25] "158062"
>> "442868" "100131137" "49" "410" "2683" [31]
>> "3010" "4184" "6677" "7142" "7455" "8857"
>> [37] "11055" "124626" "2054" "2741" "10343"
>> "10566" [43] "27297" "152015" "3074" "167"
>> "928" "2515" [49] "5104" "23553" "284359"
>> "164684" "7141" "79400"
>> Best,
>> Jim
>>
>> Tim Smith wrote:
>> Thanks James. If I can tweak that function, I'll get exactly what I
>> want. I tried what you suggested and got the following error:
>> --------------------------- ### 'genes1' are the Entrez IDs of my
>> genes of interest, and 'allGenes' is the universe of Entrez IDs
>> paramsGO <- new("GOHyperGParams", geneIds = genes1,
>> universeGeneIds = allGenes, annotation = "org.Hs.eg.db",
>> ontology = "BP", pvalueCutoff = 1, conditional = FALSE,
>> testDirection = "over") GO <- hyperGTest(paramsGO) ps <-
>> probeSetSummary(GO)
>> Error in get(mapName, envir = pkgEnv, inherits = FALSE) : variable
>> "org.Hs.egENTREZID" was not found --------------------------------
>> I guess the function would return the probe ids if I was using them,
>> but I have Entrez IDs as input.
>> Or am I doing something wrong?
>> thanks!
>>
>>
>>
>>
>> ----- Original Message ---- From: James W. MacDonald
>> <jmacdon at med.umich.edu>
>> Cc: bioc <bioconductor at stat.math.ethz.ch> Sent: Wednesday, October
>> 22, 2008 9:10:39 AM Subject: Re: [BioC] GOstat: listing genes from
>> hyperGTest
>> Hi Tim,
>> Does probeSetSummary() do what you want?
>> Best,
>> Jim
>>
>>
>> Tim Smith wrote:
>> Hi,
>> I was performing a hyperGTest for genes in homo-sapiens. For a set of
>> input genes, this function returns some 'significant' GO terms. What
>> I wanted to now do was to co-relate each significant GO term
>> (returned by this function) with genes (from my set of input genes)
>> associated with that GO term. However, I think that I may be using
>> the wrong package/function to get the releveant set of genes.
>> Currently, what I'm doing is finding the significant GO terms by
>> using the following code:
>> ----------------------- ### 'genes1' are the Entrez IDs of my genes
>> of interest, and 'allGenes' is the universe of Entrez IDs paramsGO
>> <- new("GOHyperGParams", geneIds = genes1, universeGeneIds
>> = allGenes, annotation = "org.Hs.eg.db", ontology = "BP",
>> pvalueCutoff = 1, conditional = FALSE, testDirection = "over")
>> GO <- hyperGTest(paramsGO) -------------------------- This gives me a
>> set of significant GO terms. Now, I would like to find which subset
>> of genes in 'genes1' is associated with each of the significant GO
>> term. To do this I map all GO terms to their Entrez IDs using the
>> 'org.Hs.eg.db' package using the following:
>> xx <- as.list(org.Hs.egGO2EG)
>> to get a mapping of GO terms to Entrez IDs. I get 6,756 GO terms
>> (isn't this number small?) that map to at least one Entrez ID. So,
>> from here I look up which Entrez IDs are associated with my GO term
>> of interest.
>> My problem is that often, the GO term from hyperGTest is not
>> associated with any Entrez ID (using xx <- as.list(org.Hs.egGO2EG)
>> described above ), i.e. the GO term/ID is not in the list obtained
>> from 'org.Hs.egGO2EG'). For example, the term 'GO:0043284' is thrown
>> up by hyperGTest, but does not appear to be associated with any
>> Entrez IDs in the org.Hs.eg.db package. Where could I be going wrong?
>> [[elided Yahoo spam]]
>> Thanks for any comments/suggestions. I realize that I'm probably
>> doing something really stupid here....
>> My sessionInfo() is: -------------------------------- R version 2.7.2
>> (2008-08-25) i386-pc-mingw32 locale: LC_COLLATE=English_United
>> States.1252;LC_CTYPE=English_United
>> States.1252;LC_MONETARY=English_United
>> States.1252;LC_NUMERIC=C;LC_TIME=English_United States.1252
>> attached base packages: [1] grid splines tools stats
>> graphics grDevices utils datasets methods base other
>> attached packages: [1] gplots_2.6.0 gmodels_2.14.1
>> gtools_2.4.0 gdata_2.4.1 Rgraphviz_1.18.1
>> GOstats_2.6.0 Category_2.6.0 [8] RBGL_1.16.0
>> annotate_1.18.0 xtable_1.5-2 graph_1.18.0
>> PFAM.db_2.2.0 GO.db_2.2.0 KEGG.db_2.2.0 [15]
>> org.Hs.eg.db_2.2.0 AnnotationDbi_1.2.0 RSQLite_0.6-8
>> DBI_0.2-4 genefilter_1.20.0 survival_2.34-1
>> affy_1.18.0 [22]
>> preprocessCore_1.2.0 affyio_1.8.0 Biobase_2.0.0 loaded
>> via a namespace (and not attached): [1] cluster_1.11.11 MASS_7.2-44
>> ---------------------------------
>>
>> [[alternative HTML version deleted]]
>> _______________________________________________ Bioconductor mailing
>> list Bioconductor at stat.math.ethz.ch
>> https://stat.ethz.ch/mailman/listinfo/bioconductor Search the
>> archives:
>> http://news.gmane.org/gmane.science.biology.informatics.conductor
>>
>
More information about the Bioconductor
mailing list