Hi BioC community,

I have spectral count data (proteomics) to analyze. I have two groups (A ad B) to compare with 8 patients by group. Each sample was processed in duplicate, leading to 16 distinct patients in the study and 32 samples (16 by group).

As for the method to use to detect differentially expressed proteins between the two groups, I read a lot of things in the support site and in publications, in particular :

- Use the classical edgeR with TMM normalization or DEseq (Anders et al, nature protocol, 2013)

- Use edgeR with total spectral count normalisation (like in the msms package)

- Use limma on the log2 (count+1) with quantile normalization and eBayes(trend=TRUE, robust=TRUE) : voom for spectral counts

- Use edgeR and its quasi negative binomial approach : voom for spectral counts

- PLGEM package

- ...etc

As my data are count data, I wanted to use edgeR or DEseq package. But with the post of Gordon Smyth, 5 months ago ( voom for spectral counts ), I wanted to switch to limma with quantile normalization but I have no reference that I can cite.

Did you learn more about comparing the different approach on spectral count data ? Do you have recommendations to give ? Is there an article that compare the different methods and that I can cite please ?

Thanks a lot for your help,

Eléonore

I don't have anything to add to my earlier post. As far as a reference, you could cite

  http://www.ncbi.nlm.nih.gov/pubmed/25605792

for limma or

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

for quasi-likelihood but we haven't published anything specifically about proteomics spectral data yet. It's too early for comparison papers I think.