Dear all, I am analyzing illumina data and I treat a "little" problem. If I use vst transformation as proposed by Simon Lin and Pan Du for normalization, I have a problem with interpretation of these values. I have 5 clinical groups, from which one is represents controls. If I apply log2 transformation, I can simply subtract let's say median of control group from all values and thus obtain a kind of ratio representing the relative fold change in expression. If the biologists are interested in "real", not logged expression, I can derive it using the 2^ function to the logarithm of value of relative fold change (I know this is a bit tricky, but for informative purposes still usable). But when using vst, where arcsinh is used, this task is not so simple, especially for relative fold change in comparison to control. How do you treat this problem? Thank you very much. Eva ********************************************* Eva Budinská, M.Sc. Institute of Biostatistics and Analyses Faculty of Science and Faculty of Medicine Masaryk University Kamenice 126/3, 625 00, Brno, Czech Republic ---------------------------- Tel.: +420 549 493 929 Fax.: +420 549 492 855 Mobile: +420 774 206 332 email: budinska@iba.muni.cz ********************************************* [[alternative HTML version deleted]]

Dear Eva the short answer is that you can report VST transformed data just like log-transformed data, and compute (estimates of) relative fold change as you suggest. The slightly longer answer is that asinh(x) = log(x + sqrt(x^2 + 1)) and this extra term "sqrt(x^2 + 1)" only really plays a role when x is small (for large x, asinh(x)~log(2x)=log(x)+const.) Its role is similar to what is called "background correction" in approaches that are log-transformation based, and it makes sure that (log-)ratios between small intensities are regularized towards 1 (0), in order to control (or stabilize) the variance. For example, the "normexp background correction" is a complicated non-linear transformation of the data intended to be applied before the log-transformation. Best wishes Wolfgang ---------------------------------------------------- Wolfgang Huber, EMBL-EBI, http://www.ebi.ac.uk/huber Mgr. Eva Budinsk? wrote: > Dear all, > > > > I am analyzing illumina data and I treat a "little" problem. If I use vst > transformation as proposed by Simon Lin and Pan Du for normalization, I have > a problem with interpretation of these values. > > I have 5 clinical groups, from which one is represents controls. If I apply > log2 transformation, I can simply subtract let's say median of control group > from all values and thus obtain a kind of ratio representing the relative > fold change in expression. If the biologists are interested in "real", not > logged expression, I can derive it using the 2^ function to the logarithm of > value of relative fold change (I know this is a bit tricky, but for > informative purposes still usable). > > But when using vst, where arcsinh is used, this task is not so simple, > especially for relative fold change in comparison to control. > > How do you treat this problem? > > > > Thank you very much. > > > > Eva > > > > ********************************************* > Eva Budinsk?, M.Sc. > Institute of Biostatistics and Analyses > Faculty of Science and Faculty of Medicine > Masaryk University > Kamenice 126/3, 625 00, Brno, Czech Republic > ---------------------------- > Tel.: +420 549 493 929 > Fax.: +420 549 492 855 > Mobile: +420 774 206 332 > email: budinska at iba.muni.cz > *********************************************