SEQanswers

Go Back   SEQanswers > Bioinformatics > Bioinformatics



Similar Threads
Thread Thread Starter Forum Replies Last Post
CuffDiff 0 FPKM Output/Incorrectly Identified Differential Expression? cw11 Bioinformatics 4 04-27-2015 03:59 AM
RnaSeq differential expression Gorbenzer Bioinformatics 0 10-30-2013 03:02 AM
RNA-Seq: SAMstrt: Statistical test for differential expression in single-cell transcr Newsbot! Literature Watch 0 09-03-2013 02:00 AM
RNA-Seq - From Cufflinks FPKM to differential gene expression aituka Introductions 1 07-29-2012 10:20 AM
RNAseq and Differential Expression Marianna85 RNA Sequencing 1 07-12-2012 10:44 AM

Reply
 
Thread Tools
Old 11-03-2013, 10:25 AM   #1
ndovu9
Junior Member
 
Location: Cambridge

Join Date: Jul 2013
Posts: 3
Default RNASeq Differential Gene Expression with two-sample Mann-Whitney test on FPKM?

I am interested in performing differential gene expression analysis of RNASeq data in two conditions with a number of biological replicates for each condition. I have obtained gene-level FPKMs for each sample. As gene-level FPKMs should be properly normalized, comparing the expression of a single gene using FPKM across multiple samples should be reasonable. Thus, my simple strategy is this:
use a two-sample, unpaired Mann-Whitney (Wilcox) test on the FPKMs for each gene in the two groups and correct for multiple hypothesis testing.

I have a few questions regarding this approach. First, is it valid? Second, would this approach be valid for other RNASeq normalization methods (in particular Transcripts per Million (TPM) generated by RSEM)? Third, how is this approach better or worse than either count-based methods (e.g. DEGSeq/edgeR) or other commonly used methods (e.g. Cuffdiff)?
ndovu9 is offline   Reply With Quote
Old 11-03-2013, 03:15 PM   #2
dpryan
Devon Ryan
 
Location: Freiburg, Germany

Join Date: Jul 2011
Posts: 3,480
Default

I guess it depends on how many samples you have and if the fpkm normalization works out OK or not. FPKM is effectively a tweaked library size-based normalization, with the accompanying issues. That aside, if you have a good number of samples such that you can properly use a Wilcoxon, then it's probably OK. Then biggest benefit to DESeq/edgeR/etc. is the information sharing between genes for dispersion estimation. If you really do have a lot of samples, though, I would suspect that that's not very important.
dpryan is offline   Reply With Quote
Old 01-27-2014, 01:45 AM   #3
sindrle
Senior Member
 
Location: Norway

Join Date: Aug 2013
Posts: 266
Default

"good number of samples"

How many is that? Approximatly?
sindrle is offline   Reply With Quote
Old 01-27-2014, 02:56 AM   #4
dpryan
Devon Ryan
 
Location: Freiburg, Germany

Join Date: Jul 2011
Posts: 3,480
Default

More than a few and less than a sh!t-ton*.

For real numbers you'd have to either look that up or do a quick monte-carlo (the latter is probably more useful since you would just plug in observed distributions). My pulled-from-the-hind-quarters guesstimate would be maybe 10 samples per group, since that'd be enough to get a decent sense of the distribution (but for the love of <insert random diety>, don't base anything important on that!). A quick search of the literature reveals some more modest suggestions (around 10 minimum in one group and 4 minimum in the other), though I'd have to read further to find out how robust that really is.

*interestingly, the forum replaces curse words with asterisks.
dpryan is offline   Reply With Quote
Old 01-27-2014, 02:58 AM   #5
sindrle
Senior Member
 
Location: Norway

Join Date: Aug 2013
Posts: 266
Default

Thank!
I was going to ask if 13 is ok, and I guess it is!

I would also guess 5-10 i each group..
sindrle is offline   Reply With Quote
Reply

Tags
differential expression, rnaseq, statistical analysis

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off




All times are GMT -8. The time now is 11:00 PM.


Powered by vBulletin® Version 3.8.9
Copyright ©2000 - 2020, vBulletin Solutions, Inc.
Single Sign On provided by vBSSO