normalized counts

[zhangguy]

Hi,

Thanks for depositing your code and the dismay package.

During calculating the co-expression matrices, raw umi counts was used as input, as apposed to the umi counts normalized to per cell or log(normalized umi). This is somehow different than what people would do in the bulk RNA-Seq area, where log(normCounts) are usually used. Would you worry that the co-expression metrics will be affected by using raw umi counts, which are apparently not normally distributed? Or am I missing something here?

Thanks a lot.

[suzannejin]

Hi @skinnider !! I have the same question regarding normalization. Normally normalized+log transformed bulk RNA-seq data are used for co-expression analysis. When it comes to CoDa, a centered log-ratio transformation is required. Indeed, the propr package, by default, applies CLR on the input data, and then it computes the proportionality coefficients. My question is how do you expect this to affect the benchmarking results since the other metrics (eg. stats::cor, etc) were applied to the raw data directly?