[reposting; still problems with the Google forum.] Hi Nicolas.
On Tue, Jul 20, 2010 at 11:42 AM, Nicolas Vergne <nicolas.vergne....@gmail.com> wrote: > Hi everybody, > > I use ACNE for the normalization of SNP6.0 chip arrays. > As ACNE is a multi-array methode, I would like to know if there is an > option to precise the dataset of reference in the doACNE function? You may ask one of two things. Either you want to be able (a) to specify the subset of the arrays that you trust and you wish to estimate the ACNE model parameters based on, or you wish (b) to estimate them from a separate reference (training) set. The ACNE package does unfortunately not support neither of this yet. For (a), I can only say that you have to rely on the robust estimators of ACNE and the assumption that most arrays behave as normals at any given SNP (it can be different set of samples for each SNP). For (b), the best you can do for now, is to include your training data set when you fit ACNE. If it is large enough it will dominate the estimates. As long as you do ACNE "manually" (i.e. not doACNE()): http://aroma-project.org/vignettes/ACNE you can still do the CRMAv2 preprocessing part of ACNE separately for the training data set. It is only when you get to that NmfSnpPlm step where you have to merge your test and the training data set, e.g. csNRef <- ... # Probe-normalized training data set csN <- ... # Probe-normalized test data set # Append the training (reference) set to the test data set csN <- append(csN, csNRef); # And fit the ACNE probe summarization for the lot plm <- NmfSnpPlm(csN, mergeStrands=TRUE); ... and so on. DETAILS: In order to truly use external parameter estimates ("priors"), we to be able to specify that in the NmfSnpPlm setup. Part of this mechanism is already in place (generically in the aroma.affymetrix framework), but not fully. What is mainly missing is that the internal low-level fitSnpNmf() of ACNE still don't recognize/utilize such "prior" estimates. I cannot predict when this can be done by me. You may want to look at it yourself, I recommend to get it working with fitSnpNmf(). There is an example in help(fitSnpNmfArray) that could be adjusted for testing it. When that is in place, it shouldn't be that hard for me to update NmfSnpPlm and the wrapper doACNE() accordingly. That is for alternative (b), though alternative (a) also needs to be implemented in fitSnpNmf(). > I would like to use the same sample for each new chip normalization. And > I wouldn't like to use the dataset that I want to normalize. Is it a > good way? My problem is to not reproduce the analysis for each new > chip in the project. This sounds like the (b) alternative: It is rather well known that there are large lab and batch effects in Affymetrix data. Preprocessing removes some of this but certainly not everything. Others have observed this over and over. Because of this, estimating the ACNE model parameters on one data set from a different batch/lab and use them to normalize another data set will work less well than if the parameters where estimated from samples with the same batch. Hope this helps (a bit). /Henrik > > Thanks in advance for your answers, > > Nicolas > > -- > When reporting problems on aroma.affymetrix, make sure 1) to run the latest > version of the package, 2) to report the output of sessionInfo() and > traceback(), and 3) to post a complete code example. > > > You received this message because you are subscribed to the Google Groups > "aroma.affymetrix" group with website http://www.aroma-project.org/. > To post to this group, send email to aroma-affymetrix@googlegroups.com > To unsubscribe and other options, go to http://www.aroma-project.org/forum/ > -- When reporting problems on aroma.affymetrix, make sure 1) to run the latest version of the package, 2) to report the output of sessionInfo() and traceback(), and 3) to post a complete code example. You received this message because you are subscribed to the Google Groups "aroma.affymetrix" group with website http://www.aroma-project.org/. To post to this group, send email to aroma-affymetrix@googlegroups.com To unsubscribe and other options, go to http://www.aroma-project.org/forum/
