--------------------------- Enviado con Sparrow (http://www.sparrowmailapp.com/?sig) El jueves 19 de julio de 2012 a las 10:06, Djie Tjwan Thung escribi?: > Hi Gustavo, > > Normalization, at least when you do it the "lumi" way is done at the probe level, and as each locus is assigned to a fixed color, the normalization is actually done over the M and U values. So if you have real raw data you can pull it through your pipeline: color adjustment, background subtraction, normalization. The color adjustment takes the red and green signals into account and from my understanding this is done to make the loci more comparable. That is what I was thinking about, but Tim's answers has made me think that in the 27k it might not be as important, as the probes are all of the same type. Unless, of course, we are planning to take into account the wrong channel residual intensities in more advanced methods. > > Q3) Sure that is alright. If you suspect low quality, you can use HumMeth27QCReport (a CRAN package) Thanks for the link! > to plot some of the control probe values or you can ask if they have some control plots from GenomeStudio. That's a good idea too. I'll ask them ASAP. > Ofcourse it could also be possible that the sample population is not suitable for finding DMR's (like very heterogeneous disease vs control) Yes, that might be a possibility. We currently have no clear idea of what they are trying to do, nor the global structure of their experiment. > Disclaimer: Im still a bachelor student, so dont make me accountable for any mistakes ;-) I'm sure there is someone around there with more wisdom, who can answer your questions more thoroughly. I currently hold a PhD in Machine Learning, although I have been apart from the research world for 4 years. But, when talking about bioinformatics, I am a complete beginner, willing to learn from everybody, including you, of course. :) And, if your answers are going to be as informative as this one, be sure I'll learn a lot from them. > > Cheers, > > Djie Regards, Gus > > > 2012/7/19 Gustavo Fern?ndez Bay?n <gbayon at="" gmail.com="" (mailto:gbayon="" at="" gmail.com)=""> > > Hi everybody. > > > > Imagine I have my ideal, conceptual, pipeline for working with my Illumina 27k data: > > > > A) Bead level intensities > > > > ---> (color adjustment, background subtraction, normalization, ?) > > > > ---> B) Meth and Unmeth intensities > > > > ---> (get ratios) > > > > ----> C) Beta values or M-values > > > > As I understand, proper normalization, if needed, should be done over the Red and Green signals (bead level signals, I know there are two types of beads, M and U, for each locus, and that each locus is assigned to a fixed color channel. I have studied the code in the minfi package, although it is for 450k, just to understand how these values get converted). > > > > Q1) Am I right? Or, is there any way we can normalize by using the M and U signals? > > > > Q2) As an example, I have downloaded raw data from a given GEO dataset, and found only the M and U (Signal A and B, in their nomenclature) data. At that, point, I think I cannot do neither normalization nor QC, can I? What happens if somebody tries to normalize these signals and, for example, color balance was not adjusted? Or does GenomeStudio care about that? > > > > Q3) People from another lab want us to take a look to some methylation datasets they have, and I am planning (having seen the facts above) to ask them for the .txt or .idat files, in order to do QC and normalization with lumi. Is that ok? Or is there any alternative? They claim they are having a lot of problems for finding DMR's, and I suspect they have low-quality data, and maybe that they are not following a correct pipeline. > > > > Regards, > > Gustavo > > > > --------------------------- > > Enviado con Sparrow (http://www.sparrowmailapp.com/?sig) > > > > _______________________________________________ > > Bioconductor mailing list > > Bioconductor at r-project.org (mailto:Bioconductor at r-project.org) > > https://stat.ethz.ch/mailman/listinfo/bioconductor > > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor >