Hi,
I was trying to analyze some methylation (illumina 27k) data. I have
data for 18
cancer and 18 normal samples. I want to find out if certain regions
show
consistently higher methylation in cancer (as compared to normal). A
t-test (and
then corrected for fdr) for individual probes does not reveal a
statistically
significant difference between samples.
Is there a way I can use a sliding window approach to see if there are
consistently differentially methylated regions? Is there a R and/or
bioconductor
package that does it? Or if not, which statistical methods would be
most
appropriate?
Tim
[[alternative HTML version deleted]]

Hi Tim,
the 'les' (Loci of Enhanced Significance) package could be suited for
your analysis.
In a first step you assess the effect between the groups for each
probe
with a statistical test (as you have already done with the t-test, but
you could also use other tests). Then, the les package estimates the
degree of significant probes within a sliding window along the genome.
This is based on a non-parametric estimation of the distribution of
the
p-values, and hence you do not have to specify a threshold or correct
for multiple testing. Finally, you can search for regions with a high
degree of significant probes which would correspond to methylated
regions.
We have successfully used this to analyze methylation arrays while in
principle it is suited for all kinds of tiling arrays.
Another package applying a sliding window approach is 'rMAT', but
it seems to be specific for Affymetrix tiling arrays.
Best
Julian

Hi Tim,
In the package rnaSeqMap we have implemented (also in C) the
Aumann-Lindell algorithm, which is
sort of slide-and-join algorithm on two windows across the chromosome
and
finds the regions even when there are
some small gaps. The regions found by A-L are "irreducible" -see the
paper
on theorems of this property - and thus overcome some limitations of a
single sliding window. Applying it to your data probably would require
some workaround,
but you may check rnaSeqMap vignette and paper and the original paper
of
Aumann and Lindell from 2003,
perhaps it will be of some use for you.
Cheers,
Michal
On 6/6/11 8:35 PM, "Tim Smith" <tim_smith_666 at="" yahoo.com=""> wrote:
>Hi,
>
>I was trying to analyze some methylation (illumina 27k) data. I have
data
>for 18
>cancer and 18 normal samples. I want to find out if certain regions
show
>consistently higher methylation in cancer (as compared to normal). A
>t-test (and
>then corrected for fdr) for individual probes does not reveal a
>statistically
>significant difference between samples.
>
>
>Is there a way I can use a sliding window approach to see if there
are
>consistently differentially methylated regions? Is there a R and/or
>bioconductor
>package that does it? Or if not, which statistical methods would be
most
>appropriate?
>
>Tim
>
> [[alternative HTML version deleted]]
>
>_______________________________________________
>Bioconductor mailing list
>Bioconductor at r-project.org
>https://stat.ethz.ch/mailman/listinfo/bioconductor
>Search the archives:
>http://news.gmane.org/gmane.science.biology.informatics.conductor