hi, thanks in advance I want to ask when should I use voom or trend, because I can not decide the library sizes are quite variable between samples, can you show me the code to detect this,

you once said fpkm differential analysis should use trend=T, robust =T , so should I do cpm about fpkm, if so, how should I prefilter fpkm value, is there a recommendation

and I hear someone said limma will give more false positive compared to edger and deseq2, is that true?

here is the detailed code of trend and voom If the sequencing depth is reasonably consistent across the RNA samples, then the simplest and most robust approach to dierential exis to use limma-trend. This approach will usually work well if the ratio of the largest library size to the smallest is not more than about 3-fold. In the limma-trend approach, the counts are converted to logCPM values using edgeR's cpm function:

> logCPM <- cpm(dge, log=TRUE, prior.count=3)

The prior count is used here to damp down the variances of logarithms of low counts. The logCPM values can then be used in any standard limma pipeline, using the trend=TRUE argument when running eBayes or treat. For example:

> fit <- lmFit(logCPM, design)
> fit <- eBayes(fit, trend=TRUE)
> topTable(fit, coef=ncol(design))

Or, to give more weight to fold-changes in the gene ranking, one might use:

> fit <- lmFit(logCPM, design)
> fit <- treat(fit, lfc=log2(1.2), trend=TRUE)
> topTreat(fit, coef=ncol(design))

Differential expression: voom When the library sizes are quite variable between samples, then the voom approach is theoretically more powerful than limma-trend. In this approach, the voom transformation is applied to the normalized and ltered DGEList object:

v <- voom(dge, design, plot=TRUE)

The voom transformation uses the experiment design matrix, and produces an EList object. It is also possible to give a matrix of counts directly to voom without TMM normalization, by

> v <- voom(counts, design, plot=TRUE)

If the data are very noisy, one can apply the same between-array normalization methods as would be used for microarrays, for example:

> v <- voom(counts, design, plot=TRUE, normalize="quantile")

After this, the usual limma pipelines for dierential expression can be applied, for example:

> fit <- lmFit(v, design)
> fit <- eBayes(fit)
> topTable(fit, coef=ncol(design))

Or, to give more weight to fold-changes in the ranking, one could use say:

> fit <- treat(fit, lfc=log2(1.2))
> topTreat(fit, coef=ncol(design))

Sorry, I don't understand where your difficulty lies because the text that you have copied from the limma User's Guide seems to answer your question. Indeed your post consists mostly of text copied from the User's Guide.

Are you saying that you don't know what the library sizes are for your data? Or you do know the library sizes but don't know how to decide whether one library size is more than 3 times another?

I also don't understand your references to fpkm. We strongly recommend against using fpkm for DE analyses, as you already seem to know. Even with fkpm however, limma would still be more powerful that Wilcox tests.

limma does not give more false positives than edgeR or DESeq2 and I am not aware of any publication that makes that claim.