[BioC] short time-course design. Any suggestion?
Stefano Calza
stecalza at tiscali.it
Thu Jul 22 12:18:33 CEST 2004
On Tue, Jul 20, 2004 at 07:00:37PM +0100, Adaikalavan Ramasamy wrote:
> a) Are you interested in the difference in cell lines over times OR
> b) are you treating the different cell lines as biological replicates
>
> Assuming the latter, you have a oneway anova with time as a main factor
> and 3 replicates at each time point.
That's right. Sorry, my description was not that clear. This is what I did, an ANOVA with time as a main factor, but assuming a correlation structure among observations
>
> I would suggest you try RMA and GC-RMA on the whole dataset first and
> truncating your list later. The truncation at step 2 ignores more than
> 90% of the genes and your number of true positives will be quite low.
1) Using all the genes (or most of the genes after a bit of unspecified filtering such as on the lowest expression value across samples and on the CV) brings to such a big number of comparison that
after correction none appears to be significant. Nevertheless I could use this as an exploratory approach, i.e. to rank genes.
2) Prefiltering using an "a priori" biological framework would mean (but please correct me if I'm wrong) asking a different question: among those genes related to some biological process I'm interested
in, which are actually differentially expressed?
Why shall I use RMA? E.g. with a very naive approach (i.e. computing F statistics without considering correlation among observations with arrayMagic = faster!) I get that mas5 values gives more higher F
values (a simple qqplot can help). Also the overall analysis doesn't improve using rma.
I know of affycomp but I never used it. I'll try.
Thanks.
Ste
> You can use GO tools (I think BioConductor have some packages to handle
> these) on the final gene list to see if your favourite pathway is
> involved.
>
>
>
> On Tue, 2004-07-20 at 18:17, Naomi Altman wrote:
> > You appear to have no replicates. Without replication you cannot do any
> > statistical analysis such as ANOVA or limma.
> >
> > --Naomi
> >
> > At 06:10 PM 7/19/2004 +0000, Stefano Calza wrote:
> > >Hi everybody.
> > >
> > >I'm looking at a small experiment with 12 chips (Affy), from 3 different
> > >cell lines measured at 4 different time points (0,2 hours, 8 h, 24 h).
> > >
> > >1) mas5 expression values
> > >2) selected about 1500 genes (out of ~22000) using GO annotations for
> > >those BP of possible interest
> > >3) selected genes with at least 25% Presence/Calls (I know this is quite
> > >arbitrary).
> > >4) ANOVA using gls with Compound Symmetry correlation structure
> > >5) p value corrected either using p.adjust(...,"fdr") or computing Q values.
> > >
> > >I actually get few "significant" genes and mostly with low fold-change
> > >(relative to time 0) and overall low expression intensities.
> > >Any objection about all this and/or any suggestion for improvement?
> > >
> > >Thanks in advance,
> > >Ste
> > >
> > >_______________________________________________
> > >Bioconductor mailing list
> > >Bioconductor at stat.math.ethz.ch
> > >https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
> >
> > Naomi S. Altman 814-865-3791 (voice)
> > Associate Professor
> > Bioinformatics Consulting Center
> > Dept. of Statistics 814-863-7114 (fax)
> > Penn State University 814-865-1348 (Statistics)
> > University Park, PA 16802-2111
> >
> > _______________________________________________
> > Bioconductor mailing list
> > Bioconductor at stat.math.ethz.ch
> > https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
> >
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
More information about the Bioconductor
mailing list