I'm reading the voom paper, and a piece of the notation doesn't seem right to me. voom paper link That or I don't understand the method, which is the most likely situation.

In the voom variance modeling section the following part:

In particular the following line doesn't make sense to me " a LOWESS curve is fitted to square-root standard deviations 𝑠1/2𝑔 as a function of mean log-counts 𝑅" but it seems to me we'd really like to fit the LOWESS curve as a function of r which is the mean gene level count and not R which is the geometric mean of the library size.

I think this is the case as we'd like to model variance by the mean log count of each gene and then use that to model the variance of each observation with a similar log count. I'm not sure what it would mean to model the square-root standard deviations as a function of the library size. Can someone clarify for me whats I'm missing here?

Thanks

Yes, you're quite right that the trend is a function of r rather than of R. See the corrected preprint at:

http://www.statsci.org/smyth/pubs/VoomPreprint.pdf

I tried to add a correction some years ago to the published article on Genome Biology, but the process was more difficult than I expected, so I corrected the preprint instead.