[BioC] how to deal with a complete pooling design?

[Naomi Altman]

Dear Michael,
You need to use blocks for the pair of arrays in the technical rep 
dye swaps.  If you look at the d.f. for your design, you will see 
that you have too many error d.f.
To handle the fact that all of your B samples are actually technical 
replicates, you need a mixed model that is too complicated for 
Limma.  However, you will probably be OK with the block design.

--Naomi

At 10:03 AM 9/23/2006, michael waisberg wrote:
>Hi,
>
>I have an experiment with two experimental groups. In one group there are
>three pools of at least 500 worms and in the other there is only one pool
>(the sample were pooled out of necessity - i.e lack of rna and lack of
>amplification kits).  The three pools from the first group were hybridized
>against the single pool in the second group (using a dye swap design and a
>total of 6 arrays) and therefore I have 2 df. The design is something like
>(A1 -> B,  B -> A1, A2 -> B, B -> A2, A3 -> B, B -> A3). Therefore the
>slides can be considered technical replicates (if we consider the second
>group pool in each hyb) but are not if we consider the first group pools. I
>can see that the hybs are not independent and that this is a problem but
>after reading the paper from Kedziorsky on pooling I assume that the
>accuracy of this design shouldn't be too bad since the biological
>variability between pools should be low (the number of subjects in each pool
>is high) in comparison to the technical variability. I analysed the data
>using limma with a design matrix:
>
>(1,-1,1,-1,1,-1)
>
>Where 1 and -1 are the dye swaps. Can I trust (at least in part) the p and B
>values generated from limma? If not, what kind of criteria can I use to
>decide which genes are relevant? Do I have to limit my analysis to fold
>change? Can I trust the ranking of genes generated by limma with this
>analysis?
>
>I tought about using the variation between the technical replicates as my
>best guess for the variation across samples, performing the analysis using
>these technical replicates and then using the samples witgh B values >
>4.6and fold change > 2 as my criteria for selecting "significant"
>genes. Is
>this reasonable?
>
>Best regards,
>
>Michael Waisberg
>
>         [[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

Naomi S. Altman                                814-865-3791 (voice)
Associate Professor
Dept. of Statistics                              814-863-7114 (fax)
Penn State University                         814-865-1348 (Statistics)
University Park, PA 16802-2111