On Fri, Aug 21, 2009 at 12:27 PM, Tim Triche <tim.triche@gmail.com> wrote: > watch out for > > 1) dye bias (R/G) > 2) probe bias (final base before the interrogated site, at least) > The methylumi package normalizes the dye bias by looking at nearly-fully-methylated and nearly-fully-unmethylated probes, but we do not deal with probe bias as of yet. It works with GG and Infinium data, but we have not dealt with custom arrays directly. Sean > > for Infinium arrays there has been some discussion as to whether quantile > normalization squashes subtle biological differences, thus exploration of > #2. That's considering that the dye bias is mostly removed from play. > With > GG methylation arrays, the smaller number of probes (especially a custom > array!) and presence of dye bias suggests treading carefully. If you have > replicates among your data, try looking at the effects of affine, quantile, > and PLM/BLM normalization between each. I'd compare notes, but we're on > Infinium. > > > I personally have been working with twin data on the Infinium arrays, along > with a high number of replicates from a single whole blood sample (my P.I. > got burned by a subtle Illumina omission when his first data set was run). > While the noise from identical samples is much smaller than that within > twin > pairs and between unrelated individuals or different cell types, it's > there, > and may be significant enough (even with just 384 targeted probes, as > yours) > to consider normalization. Again, my data is unusual in that we are more > interested in differences within pairs than anything else, but for > permutation testing I do want a common baseline. So we are pursuing > appropriate normalization strategies. > > If memory serves, Roche/Nimblegen proudly notes that they do not normalize > their methylation array results, so there's another perspective. > > Hope this helps, > > --t > > > On Fri, Aug 21, 2009 at 4:20 AM, r.kandimalla <r.kandimalla@erasmusmc.nl> >wrote: > > > Dear all, > > > > I would like to know some information regarding normalisation and > > statistics to apply on my custom GGMA assays. > > > > Is normalisation necessary for these assays ? if yes what do you suggest > ? > > > > I havent applied any statistics on the data to find differential > > methylation among subgroups, instead i just took the beta values and did > > comparisons. To elaborate, i have two different subgroups of tumors in my > > data set. To compare them i just took the ratio of avg beta of one sub > > groups to avg beta of another subgroup and came up with significant list > of > > genes (i have chosen the ratio > 2 to be significant in that particular > > comparison). > > > > This was a validation assay of our genome wide screen using agilent > arrays. > > With custom GGMA consisting of 384 probes, we were able to validate quiet > > some data and is encouraging, but i have questions regarding the analysis > of > > ggma data, whether im doing something wrong ?? > > > > Your input is highly appreciated. > > > > Best regards, > > > > Raju > > > > > > -- > > Raju kandimalla, PhD student > > Department of Pathology > > Josephine Nefkens Institute > > Erasmus MC, Be-302 > > P.O. Box 2040 > > 3000 CA Rotterdam > > The Netherlands > > Tel: +31-10-7043093 > > > > _______________________________________________ > > Bioconductor mailing list > > Bioconductor@stat.math.ethz.ch > > https://stat.ethz.ch/mailman/listinfo/bioconductor > > Search the archives: > > http://news.gmane.org/gmane.science.biology.informatics.conductor > > > > > > -- > I'm not convinced that faith can move mountains, > but I've seen what it can do to skyscrapers. > > ( from Bill Gascoyne, http://tinyurl.com/mfwwuw ) > > [[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor@stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: > http://news.gmane.org/gmane.science.biology.informatics.conductor > [[alternative HTML version deleted]]