Hello everyone, My lab is shortly going to run a large cohort of samples (2 X 96 well plates) on a Goldengate Methylation Array. I have been playing with the data produced from a pilot study on BeadStudio 3.0 but am finding that the software has some limitations, specifically relating to QC. >From my reading, it seems that beadarray would be a good choice for carrying out more in-depth QC. I have used R before so am capable of creating a simple script to crunch through QC on all the samples, so I have a few questions for the experts: Do I have to spoof beadarray to think that it is importing expression data? If so, what are the transformations I need to do? The actual experiment is going to be run at our collaborator's site. Am I right that for us to receive raw data from BeadScan suitable for beadarray analysis, the settings.xml file must be altered to output .txt files. Finally, does anyone have a feel for an appropriate normalization technique for this type of data (it returns a beta score which has a range of 0 to 1, corresponding from fully unmethylated to fully methylated respectively)? BeadStudio has average (which minimizes variation across SAMs) and background (removes outliers using a MAD method) options in a differential methylation analysis. Thanks for reading, and I'd appreciate any comments you might have. Ed Schwalbe

Hello Ed, I think the best option would be work from the raw (bead level) data when trying to understand more about the technology. To do this, the settings.xml file will indeed need to be modified prior to scanning. I can give you more details about that if you're not sure how to do it. The function readIllumina is designed to work with the bead level data from any Illumina assay, so there is no modification needed to make it work for methylation. This will give you a red and green intensity for each bead. After QC on the bead level data, you can then choose to summarise the replicate red and green values for each bead type (createBeadSummaryData) or get an average log-ratio for each bead type. I haven't yet implemented a method to get an average beta value for each bead type. The issue of normalisation is a tricky one and I haven't seen a good solution yet unfortunately. Unlike gene expression, you can't assume most probes will not be differentially expressed or that the distribution of signal on each array will be the same. That seems to rule out some popular approaches. From the data I have seen, there is an obvious dye-bias in the two channels, which could seriously affect if you call things as being methylated or not. I think the best you can do at this stage is to normalise for this, possibly using information from the negative controls. I'd also be interested to hear if anyone has any good suggestions though! Hope this helps, Mark On 20 Nov 2007, at 16:00, Ed Schwalbe wrote: > Hello everyone, > > > My lab is shortly going to run a large cohort of samples (2 X 96 well > plates) on a Goldengate Methylation Array. I have been playing with > the data produced from a pilot study on BeadStudio 3.0 but am finding > that the software has some limitations, specifically relating to QC. > >> From my reading, it seems that beadarray would be a good choice for > carrying out more in-depth QC. I have used R before so am capable of > creating a simple script to crunch through QC on all the samples, so I > have a few questions for the experts: > > Do I have to spoof beadarray to think that it is importing expression > data? If so, what are the transformations I need to do? > The actual experiment is going to be run at our collaborator's site. > Am I right that for us to receive raw data from BeadScan suitable for > beadarray analysis, the settings.xml file must be altered to output > .txt files. > > Finally, does anyone have a feel for an appropriate normalization > technique for this type of data (it returns a beta score which has a > range of 0 to 1, corresponding from fully unmethylated to fully > methylated respectively)? BeadStudio has average (which minimizes > variation across SAMs) and background (removes outliers using a MAD > method) options in a differential methylation analysis. > > Thanks for reading, and I'd appreciate any comments you might have. > > Ed Schwalbe > > _______________________________________________ > 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