Normalization and analysis of proteomic spectral count data

bjackson · 05-08-2015, 10:37 AM

I have spectral count data (SomaScan proteomics read).

What I have is protein name / uniprot identifier and a count (e.g., 188.9 or 106219.5).

I wanted to use PLGEM in R / bioconductor to analyze the data for differential expression between groups, but it says that 'normalized' data is required. Does anyone have a recommendation of what type of normalization to do?

I can't do something like dNSAF since I only have counts for called proteins, not counts for peptides.

Thanks.

Update: I've seen two thoughts on normalization, horizontal and vertical
one would be to normalize across samples (eg divide all protein counts by the total protein count for that sample so that that normalizes to the total protein load)

The other would be to normalize across each protein (e.g. mean 0, sd 1, so highly expressed proteins don't have an outsized effect on the significance levels)

You could even do the first then the second. Is there any good reason to to either or both?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
RNA seq data filter low count before or after normalization	qliu	RNA Sequencing	4	01-10-2015 02:21 AM
Different normalization methods with count data	greener	Bioinformatics	2	09-01-2013 06:27 AM
ChIP-Seq: A normalization strategy for comparing tag count data.	Newsbot!	Literature Watch	0	04-06-2012 06:00 AM
ChIP-Seq Data Analysis and Normalization	snape_ar	Bioinformatics	0	10-25-2011 01:16 PM

05-08-2015, 10:37 AM	#1
bjackson Junior Member Location: Denver, CO Join Date: May 2015 Posts: 6	Normalization and analysis of proteomic spectral count data I have spectral count data (SomaScan proteomics read). What I have is protein name / uniprot identifier and a count (e.g., 188.9 or 106219.5). I wanted to use PLGEM in R / bioconductor to analyze the data for differential expression between groups, but it says that 'normalized' data is required. Does anyone have a recommendation of what type of normalization to do? I can't do something like dNSAF since I only have counts for called proteins, not counts for peptides. Thanks. Update: I've seen two thoughts on normalization, horizontal and vertical one would be to normalize across samples (eg divide all protein counts by the total protein count for that sample so that that normalizes to the total protein load) The other would be to normalize across each protein (e.g. mean 0, sd 1, so highly expressed proteins don't have an outsized effect on the significance levels) You could even do the first then the second. Is there any good reason to to either or both? Last edited by bjackson; 05-08-2015 at 02:45 PM.