Can I test for differential expression using FPKM values?

JonB · 11-25-2013, 05:34 AM

Hi,

I know this subject has been discussed in many posts, and I am sorry if I am double posting.

I have a matrix of gene expression counts and the values are FPKM.
I don't have the sam files.

Can I use cuffdiff with this input?

I can run DESeq2 the way it is described in the tutorial, but it says that the input should be raw counts. What happens if the input is FPKM?

Are there other methods to run DE-test on FPKM values?

Thanks,

Jon

dpryan · 11-25-2013, 06:17 AM

If you input FPKMs into DESeq2 you'll get unreliable results, since you're violating the statistical premise. You can probably use voom() to get things ready for analysis in limma. That's what I'd try first.

JonB · 11-25-2013, 06:29 AM

Ok, thanks!

JonB · 11-26-2013, 02:18 AM

Sorry for spamming you with questions...

I am able to convert the read counts by voom(), but I have trouble getting the experiment design correct using model.matrix()

I have 18 libraries representing two different conditions: 'normal' and 'post-inversion'. My design looks like this now:
design
I think I have to create a factor with the two labels 'normal' and 'post-inversion'?

JonB · 11-26-2013, 02:38 AM

And now I just realized there's a bioconductor mailing list which is probably where this question belong

dpryan · 11-26-2013, 02:46 AM

Yeah, you can more easily ask the authors on the list

If you're just interested in the normal vs. post-inversion comparison (rather than also looking at whatever that "Sycon2" thing is), then yes, just creating a factor with those levels (design$libType will provide that, though you probably named the dataframe something else).

Simon Anders · 11-27-2013, 11:51 PM

Just for the record: "voom" expects counts the same as edgeR and DESeq do. See also https://stat.ethz.ch/pipermail/bioco...er/056309.html

atulchandra · 03-05-2018, 03:11 AM

Could you tell me anyone, why we use TPM, FPKM or RPKM normalization in single-cell expression analysis?

Is there any biological explanation????

Thanks in advance.!

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
CuffDiff 0 FPKM Output/Incorrectly Identified Differential Expression?	cw11	Bioinformatics	4	04-27-2015 04:59 AM
RNASeq Differential Gene Expression with two-sample Mann-Whitney test on FPKM?	ndovu9	Bioinformatics	4	01-27-2014 03:58 AM
RNA-Seq: SAMstrt: Statistical test for differential expression in single-cell transcr	Newsbot!	Literature Watch	0	09-03-2013 03:00 AM
Cuffdiff FPKM and test statistic calculations	PRingler	RNA Sequencing	2	10-16-2012 03:47 AM
RNA-Seq - From Cufflinks FPKM to differential gene expression	aituka	Introductions	1	07-29-2012 11:20 AM

11-25-2013, 05:34 AM	#1
JonB Member Location: Norway Join Date: Jan 2010 Posts: 83	Can I test for differential expression using FPKM values? Hi, I know this subject has been discussed in many posts, and I am sorry if I am double posting. I have a matrix of gene expression counts and the values are FPKM. I don't have the sam files. Can I use cuffdiff with this input? I can run DESeq2 the way it is described in the tutorial, but it says that the input should be raw counts. What happens if the input is FPKM? Are there other methods to run DE-test on FPKM values? Thanks, Jon

11-25-2013, 06:17 AM	#2
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,480	If you input FPKMs into DESeq2 you'll get unreliable results, since you're violating the statistical premise. You can probably use voom() to get things ready for analysis in limma. That's what I'd try first.

11-25-2013, 06:29 AM	#3
JonB Member Location: Norway Join Date: Jan 2010 Posts: 83	Ok, thanks!

11-26-2013, 02:18 AM	#4
JonB Member Location: Norway Join Date: Jan 2010 Posts: 83	Sorry for spamming you with questions... I am able to convert the read counts by voom(), but I have trouble getting the experiment design correct using model.matrix() I have 18 libraries representing two different conditions: 'normal' and 'post-inversion'. My design looks like this now: design I think I have to create a factor with the two labels 'normal' and 'post-inversion'?

11-26-2013, 02:38 AM	#5
JonB Member Location: Norway Join Date: Jan 2010 Posts: 83	And now I just realized there's a bioconductor mailing list which is probably where this question belong

11-26-2013, 02:46 AM	#6
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,480	Yeah, you can more easily ask the authors on the list If you're just interested in the normal vs. post-inversion comparison (rather than also looking at whatever that "Sycon2" thing is), then yes, just creating a factor with those levels (design$libType will provide that, though you probably named the dataframe something else).

11-27-2013, 11:51 PM	#7
Simon Anders Senior Member Location: Heidelberg, Germany Join Date: Feb 2010 Posts: 994	Just for the record: "voom" expects counts the same as edgeR and DESeq do. See also https://stat.ethz.ch/pipermail/bioco...er/056309.html

03-05-2018, 03:11 AM	#8
atulchandra Junior Member Location: Japan Join Date: Mar 2018 Posts: 1	Could you tell me anyone, why we use TPM, FPKM or RPKM normalization in single-cell expression analysis? Is there any biological explanation???? Thanks in advance.!