[BioC] very small fold change from RMA data

[w.huber at dkfz-heidelberg.de]

Hi Dapeng Cui,

there is a trade-off between variance and bias when estimating
fold-change, esp. for weakly expressed genes. "Bias" means that the
estimated fold-changes for genes whose true fold-changes are different
from 1 are shrunken towards 1.

While the MAS-software stays on the high-variance, small-bias side, RMA
tends to sacrifice some bias for a large gain in variance. Thus, the
fold-changes from RMA will often be closer to 1 than those from MAS, esp.
for weakly expressed genes.

Best regards

-------------------------------------
Wolfgang Huber
Division of Molecular Genome Analysis
German Cancer Research Center
Heidelberg, Germany
Phone: +49 6221 424709
Fax:   +49 6221 42524709
Http:  www.dkfz.de/abt0840/whuber
-------------------------------------

On Thu, 16 Oct 2003, Dapeng Cui wrote:

> Hi,
>
> My question is about very small fold change from RMA treated data. What
> I did is to normalize affy chips with RMA, transform to linear values
> then import to GeneSpring. Data was filtered using Affy detection call.
> Then a two group T-Test (10 replicates in each conditions) was performed
> and gave me a list of 300 gens.  What makes me confused is that 50% of
> these genes have fold changes less than 1.2, only 10% genes have higher
> than 1.5 fold change.  Is this normal?
>
> I also did the same T-Test with Affy pre-scaled data (and nomarlized
> with GeneSpring per chip and per gene median normalization). This time
> 50% of genes in the list have higher than 1.5 fold changes. (And this
> list is not the same as that one from RMA data, only 70% overlapping.)
>
> I will appreciate it very much if anybody could provide me some literature regarding different normarlizations and fold change. Thanks.
>
> dapeng cui
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>