vsn on log2 transformed data?
2
0
Entering edit mode
@knaxerovixurzuni-heidelbergde-1612
Last seen 9.6 years ago
Hi! Could anyone explain to me *why* I cannot apply vsn to log2 transformed data? "...may not satisfy the requirements of the multiplicative-additive noise model" is a little hard to grasp for someone without a rigorous statistical background and the vignette and the original publication did not enlighten me very much. Does this really mean I cannot use vsn on a data set that contains log2 transformed ratios from a dual channel experiment? Thanks for any help!!! Kamila
vsn vsn • 1.3k views
ADD COMMENT
0
Entering edit mode
@james-w-macdonald-5106
Last seen 4 hours ago
United States
Hi Kamila, Kamila Naxerova wrote: > Hi! > > Could anyone explain to me *why* I cannot apply vsn to log2 transformed > data? "...may not satisfy the requirements of the > multiplicative-additive noise model" is a little hard to grasp for > someone without a rigorous statistical background and the vignette and > the original publication did not enlighten me very much. Does this > really mean I cannot use vsn on a data set that contains log2 > transformed ratios from a dual channel experiment? One of the reasons people log transform their microarray data is to decrease (or eliminate) the dependence between the mean and variance. In other words, with un-logged data, the variance between samples will tend to change depending on the intensity of the spot. Many statistical tests assume independence between the mean and variance, so we want to minimize the dependence as best as possible. vsn is also designed eliminate the dependence between the mean and variance. It makes different assumptions about the structure of the dependence (the multiplicative-additive noise model you mention above), but in the end the goal is the same as taking logs. Now, if you have log2 transformed data, you have _already_ done something to eliminate the mean-variance dependence. These data most likely won't "satisfy the requirements of the multiplicative-additive noise model", because that model assumes the data are on the original scale, and that the variance is still dependent on the mean. Since this is not true, you may not get reasonable results, and it is really not necessary. Long story short, if you want to use vsn on your data, you need to start with the background-corrected un-logged cy3 and cy5 intensity values. Does that help? Best, Jim > > Thanks for any help!!! > Kamila > > _______________________________________________ > 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 -- James W. MacDonald University of Michigan Affymetrix and cDNA Microarray Core 1500 E Medical Center Drive Ann Arbor MI 48109 734-647-5623 ********************************************************** Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues.
ADD COMMENT
0
Entering edit mode
@wolfgang-huber-3550
Last seen 3 months ago
EMBL European Molecular Biology Laborat…
Kamila Naxerova wrote: > Hi! > > Could anyone explain to me *why* I cannot apply vsn to log2 transformed > data? "...may not satisfy the requirements of the > multiplicative-additive noise model" is a little hard to grasp for > someone without a rigorous statistical background and the vignette and > the original publication did not enlighten me very much. Does this > really mean I cannot use vsn on a data set that contains log2 > transformed ratios from a dual channel experiment? > > Thanks for any help!!! > Kamila Hi Kamila, it is the same reason why you wouldn't apply the log transformation again on log transformed data, as James has already pointed out. This question has come up before, and I've even seen this done in published papers: please, please, users, don't call vsn on log-transformed data or on log-ratios, it is non-sensical. Note that for many practical purposes (if you don't care about maths), vsn is similar to x -> log( s*(x+x0) ) and what vsn has is a good way of estimating the parameters s and x0 (which may be negative). Many other people call this simply "background correction". If you look at two-color data, s and x0 are estimated separately for each channel, the data are transformed, then the difference is taken. If you want to know what vsn does exactly, it is the transformation x -> log ( s/2*x + s/2*sqrt ( x^2 + x0^2 ) ) this results from the additive-multiplicative error model (which is arguably the most simple error model that fits to microarray data) Hope this helps Wolfgang ------------------------------------------------------------------ Wolfgang Huber EBI/EMBL Cambridge UK http://www.ebi.ac.uk/huber
ADD COMMENT

Login before adding your answer.

Traffic: 595 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6