Entering edit mode
Hi Shilin,
The confusion is in the groups. Data columns 1-4 are paired with
columns
5-8 respectively, as you have identified. However, there is a further
structure to the data, which is that the first two paired samples are
biologically distinct from the second two paired samples. Replicate
group 1 is thus 159:73, 44:24, and replicate group 2 is 0:49, 0:68 -
the
groups are defined on the *paired* data. This situation might arise,
for
example, if the first two pairings consist of normal and tumour tissue
from patients responding to treatment, and the second two normal and
tumour tissue patients from non-responders. The replicate structure
'c(1,1,2,2)' and the DE group is designed to describe this structure -
the first two pairings are distinct from the second two pairings.
The DE analysis (correctly) identifies that there is a large and
consistent difference in ratio between the two groups - in the first
group, the data are 159:73 and 44:24, so the first member of each pair
is approximately twice that of the second member; in the second group,
the data are 0:49 and 0:68, so the first member of each pair is
substantially less than that of the second member.
If the question you would like to answer is 'where are there
consistent
differences between the paired data', rather than 'where are there
consistent differences in the ratios of paired data between replicate
groups', you can use the nullProps = 0.5 option as given in the
vignette, and look at the topCounts table for the non-differentially
expressed (between replicates) data (page 9 of the vignette).
Hope that helps,
Tom
>
> ------- Original Message --------
> Subject: [BioC] Paired Data Analysis in baySeq package
> Date: Wed, 12 Feb 2014 00:14:44 -0600
> From: zhao shilin <zhaoshilin@gmail.com>
> To: bioconductor@r-project.org <bioconductor@r-project.org>
>
>
>
> Hi all,
>
> Is there anybody who is familiar with the Paired Data Analysis in
baySeq
> package. I'm following the instructions in its vignette. I used:
>
> library(baySeq)
> data(pairData)
> pairCD <- new("pairedData", data = pairData[,1:4], pairData =
> pairData[,5:8],replicates = c(1,1,2,2),groups = list(NDE =
c(1,1,1,1), DE =
> c(1,1,2,2)))
>
> As it indicated, The first four columns in these data are paired
with the
> second four columns. So I think Sample 1-Sample 4 is group1 and
Sample
> 5-Sample 8 is group2. And Sample 1 is paired with Sample 5, Sample 2
is
> paired with Sample 6~~~
> In the result, the most significant gene is the 5th row. The result
is:
>
> rowID X1.1 X1.2 X2.1 X2.2 Likelihood DE FDR.DE
> 1 5 159:73 44:24 0:49 0:68 0.9974276 1>2 0.002572417
>
> Its expression is:
> 159 44 0 0 73 24 49 68
>
> It is very obvious that the software take 3th and 4 th samples (0,0)
as
> group 2, 7th and 8th samples (49, 68) as group 1, which is not
correct.
> So I am not very clear with the replicates = c(1,1,2,2) and DE =
> c(1,1,2,2). What do they mean here? What is the correct method to do
paired
> data analysis in baySeq package?
>
> Thank you!
>
> Best,
> Shilin
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives:http://news.gmane.org/gmane.science.biology.info
rmatics.conductor
>
>
--
Dr. Thomas J. Hardcastle
Department of Plant Sciences
University of Cambridge
Downing Street
Cambridge, CB2 3EA
United Kingdom
[[alternative HTML version deleted]]
