[BioC] extracting significant genes using limma

Stefano Calza stecalza at tiscali.it
Mon Mar 13 17:24:08 CET 2006


On Mon, Mar 13, 2006 at 10:41:08AM -0500, Naomi Altman wrote:
<Naomi>Since you used "adjust=fdr", the p-value column of the TopTable are 
<Naomi>the "adjusted p-values" after fdr (which I think of as q-values).

I'm not that sure about this.

Reading the code of toptable it uses p.adjust(...,adjust.method="fdr"), which strictly is not exactly a q-value. You get the same if 
p0 = 1, which may not be the case.

Using qvalue(toptable$P.Value) will give you a q-value (according to Storey et al.)

HIH,
Ste


<Naomi>
<Naomi>You can either pick some q-value you want to use to select the 
<Naomi>significantly differentially expressing genes, or you can pick some 
<Naomi>number of genes, and report the q-value of the least significant of these.
<Naomi>
<Naomi>--Naomi
<Naomi>
<Naomi>At 10:17 AM 3/13/2006, Assa Yeroslaviz wrote:
<Naomi>>Hi,
<Naomi>>
<Naomi>>I know this theme is an old one, but I look all over the archives and didn't
<Naomi>>find any help regarding this subject.
<Naomi>>Using Affymetrix chips I compared two groups (Control vs compound) with the
<Naomi>>limma procedure.
<Naomi>>I made an affybatch Object using ReadAffy(), normalised the data with the
<Naomi>>RMA algorithm and fitted a linear model with lmFit.
<Naomi>>
<Naomi>> >affy <- ReadAffy(filenames=vec)
<Naomi>> >eset <- rma(affy)
<Naomi>> >design <- cbind(Control=1,AE0627vsCT=c(rep(0,6),rep(1,4)))
<Naomi>>
<Naomi>>my design matrix looks like that (I have 6 control  and 4 treated arrays):
<Naomi>> > design
<Naomi>>       Control AE143vsCT
<Naomi>>  [1,]       1         0
<Naomi>>  [2,]       1         0
<Naomi>>  [3,]       1         0
<Naomi>>  [4,]       1         0
<Naomi>>  [5,]       1         0
<Naomi>>  [6,]       1         0
<Naomi>>  [7,]       1         1
<Naomi>>  [8,]       1         1
<Naomi>>  [9,]       1         1
<Naomi>>[10,]       1         1
<Naomi>>
<Naomi>>so I don't need any contrast matrix.
<Naomi>>The list is 22,810 genes long. But not all of them can be significant. I
<Naomi>>hope!!!
<Naomi>>
<Naomi>>I sorted the genes with:
<Naomi>> >sig_table <- topTable(fit_e, coef=2, number=6000, adjust="fdr", sort.by=
<Naomi>>"P")
<Naomi>>
<Naomi>>I've chosen 6000 as an arbitrary value, but I still don't know how many
<Naomi>>genes are siginificant.
<Naomi>>
<Naomi>>My question(s) is(are):
<Naomi>>
<Naomi>>1. How do I find out how many genes are significantly differentially
<Naomi>>expressed using a define p-value or FDR?
<Naomi>>     Can I use here the decideTests() function although I don't have any
<Naomi>>contrasts?
<Naomi>>
<Naomi>>2. In SAM one can look for the false discovery rates using the different
<Naomi>>delta-values.
<Naomi>>    Is it possible to set a fixed FDR-Value in Limma?
<Naomi>>    Where Do I find the FDR rates of my significant genes?
<Naomi>>
<Naomi>>3. Is there a possibility (like in SAM) to show the results in a graphic (
<Naomi>>scatter plot etc.)?
<Naomi>>
<Naomi>>Every comment and suggestion would be appreciated!
<Naomi>>
<Naomi>>THX
<Naomi>>
<Naomi>>Assa
<Naomi>>
<Naomi>>--
<Naomi>>Assa Yeroslaviz
<Naomi>>Loetzener Str. 15
<Naomi>>51373 Leverkusen
<Naomi>>
<Naomi>>         [[alternative HTML version deleted]]
<Naomi>>
<Naomi>>_______________________________________________
<Naomi>>Bioconductor mailing list
<Naomi>>Bioconductor at stat.math.ethz.ch
<Naomi>>https://stat.ethz.ch/mailman/listinfo/bioconductor
<Naomi>
<Naomi>Naomi S. Altman                                814-865-3791 (voice)
<Naomi>Associate Professor
<Naomi>Dept. of Statistics                              814-863-7114 (fax)
<Naomi>Penn State University                         814-865-1348 (Statistics)
<Naomi>University Park, PA 16802-2111
<Naomi>
<Naomi>_______________________________________________
<Naomi>Bioconductor mailing list
<Naomi>Bioconductor at stat.math.ethz.ch
<Naomi>https://stat.ethz.ch/mailman/listinfo/bioconductor

-- 
Stefano Calza, PhD
Researcher - Biostatistician
Sezione di Statistica Medica e Biometria
Dipartimento di Scienze Biomediche e Biotecnologie
Università degli Studi di Brescia - Italy
Viale Europa, 11 25123 Brescia
email: calza at med.unibs.it
Phone: +390303717653
Fax: +390303717488



More information about the Bioconductor mailing list