Entering edit mode
if I use deseq or edger or limma to do count differential analysis, what kind of data should be sent to do cox km lasso and so on downstream analysis, cpm(x), vst(x), rlog(x), voom(x) or anything else.
thanks a lot
thanks a lot. do you know edger, does it use cpm is ok? do you know limma, does it use voom is ok?
The suggested way to use the support site if you are sending notifications to package developers, is to ask help on their package specifically. I only have time to provide support for my package. If you have a question about another package, make a new post and tag only that package.
thanks a lot. so if I want to just one gene difference in two groups(for example show a differential gene analysed by deseq2), I see people always use tpm or fpkm, but not vst(dds), how to understand it,
Hi linouhao, please use either VST if using DESeq2, or CPM (with a prior count) if using EdgeR or limma-voom. For further information on distinguishing between EdgeR and limma-voom, please see here: https://support.bioconductor.org/p/59846/#59917
Please do not use FPKM units for a Cox model. FPKM units are not suitable for anything where cross-sample comparisons are to be performed.
CPM (with a prior count) ， how to set the prior count, just 1 or 2 or other int number ,and should cpm use the log=T or not， this cpm value seems just using for filter or downstream analysis other than DE, so I just use cpm(x, log=T, prior.count=1) apply for all, is it ok? thanks a lot
Hey, for Cox models, yes, I would use
log = TRUE. Regarding the prior count, I would use the default and then check the data distribution via a histogram. Basically, for a Cox model, we want a data distribution that will result in normally-distributed residuals after the model has been fit to the data (so, a good starting point is that the data distribution itself is normal - think 'bell curve').We could not use FPKM units or RNA-seq count data for a Cox due to the fact that these are not measured on a normal distribution.
thanks a lot. if I wan to use TCGA data as train data, and geo as validation, one is counts data, one is microarray, how should I do?
If you are using one dataset as a training and the other as a validation, it technically does not matter that they are from different sources (microarray versus RNA-seq), as they will be analysed independently. Same is true if you are just aiming to replicate results across the 2 studies. However, if you want integrate these datasets, then that requires further thought.
Anyway, that is a more general question, so, please ask on a forum for general Bioinformatics, e.g., Biostars (I am also Moderator on Biostars). For the record: I am sure that I have already answered questions on this topic on Biostars, if you can search via your search engine of choice.
but when you need to make a model, for example, a model contains several gene expression, you will find the editor will ask you the data comes from different platform, can it use the same fomula? thanks a lot
I am unsure what you mean and you have provided no examples.