Hello,

I have a question about the use of voomWithQualityWeights.

Is it appropriate to use this function (as apposed to ordinary voom) when the there is variability in the sample-specific weights between sample groups of interest?

For example, consider a situation where samples were collected from two tissues, tissue A and tissue B, both before and after a drug treatment, and we want to use makeContrasts and contrasts.fit to extract some contrasts of interest:

Tissue A post-treatment vs Tissue A pre-treatment
Tissue B post-treatment vs Tissue B pre-treatment
(Tissue A post-treatment vs Tissue A pre-treatment) vs (Tissue B post-treatment vs Tissue B pre-treatment)

If the sample-specific weights were lower in tissue A than tissue B, i.e. the "low quality" samples were concentrated in tissue A, would this be the sort of situation where ordinary voom would be more appropriate than voomWithQualityWeights?

In this situation might the variability in sample weights between tissues represent biological variability and not variation in quality due to technical factors?

Thanks very much

I would recommend now for you to use voomLmFit(), which has all the functionality of voomWithQualityWeights() but with some more features and more automation. Just set sample.weights=TRUE.

Yes, it is fine to use sample weights that vary systematically between groups. In fact, one of the purposes of sample weights is to allow for heteroscedasticity between groups, where one group (such as cancer) is biologically more variable than another (like normal or wild type). The weights represent variability from all sources, both technical and biological. voomWithQualityWeights() and voomLmFit() even have an argument, var.group, that allows you to make the weights specifically depend on the group.