[BioC] Finding differentially expressed genes using Ebayes

[James W. MacDonald]

Hi John,

John Herbert wrote:
> Hello.
> I have 6 gpr files, 3 from cancer and 3 from normals. I have normalised them with printtiplowess and removed the control spots. My question is about finding differentially expressed genes with Ebayes. I have run the code below and get some strange results that are very different to using Samr. I am thinking I trust samr more as when I look at the results by eye, they seem to order the genes coffectly based on fold ratio. However, it could be that I am not using ebayes correctly. 
> 
> So, the design is supposed to represent cancer (the 1s) and normal (the 2s). Is this the correct way to construct a design matrix? I want to compare gene expression between cancer and normal (3 replicates each).  
> 
> design <- model.matrix(~factor(c(1,1,1,2,2,2)));
> 
> fit <- lmFit(probes_only.imp, design);
> 
> fit <- eBayes(fit);
> 
> probes_only.imp_50_ebayes <- topTable(fit, n=20);

The model you are fitting has an intercept, which in this case measures 
the average expression of the cancer genes. So your topTable will just 
give those genes that have an average expression different from zero (an 
uninteresting result).

If you do topTable(fit, coef = 2), you will get the genes that are 
different between cancer and normal, but the direction might be 
different than what you expect (the second coefficient is the difference 
between normal and cancer, e.g., normal - cancer), so a negative result 
indicates upregulation in cancer. This might be a bit counterintuitive, 
so you could simply replace your call to lmFit() with

fit <- lmFit(~factor(rep(1:2, each = 3), levels=2:1)

which will re-order things so coefficient 2 will be cancer - normal.

Best,

Jim


> 
> Any help appreciated on this, thank you. 
> 
> Kind regards,
> 
> John.
> 
> 
> =================================
> Bioinformatics Officer
> 
> Molecular Angiogenesis group
> First Floor, Institute of Biomedical Research
> The University of Birmingham
> Medical School
> Edgbaston
> Birmingham
> B15 2TT  (UK)
> 
> j.m.herbert at bham.ac.uk
> 
> Tel:  +44 121 414 3733
> Fax: +44 121 415 8677 
> 
> _______________________________________________
> 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

-- 
James W. MacDonald, M.S.
Biostatistician
Douglas Lab
University of Michigan
Department of Human Genetics
5912 Buhl
1241 E. Catherine St.
Ann Arbor MI 48109-5618
734-615-7826