Can I apply DESeq to techinical replicates??

Sora Yoon · 04-16-2015, 01:47 AM

Hi guys,

I've been dealing with RNA-seq count data consisting of several technical replicates.
And I tried to normalize the data using DESeq.
However, the DESeq author suggested to apply DESeq to biological replicates not to technical replicates.
Anyone knows the reason for this??
Why can't I apply DESeq normalization to my technical replicates????

Also, is there any method to normalize a technical replicates effectively??

Thank you for reading.

bastianwur · 04-16-2015, 03:03 AM

The normalization is meant to take the biological variations into account, IIRC.
The variance in the technical replicates will be lesser than in biological replicates, therefore the normalization will not have the right impact (less, I'd guess).

If you're sure that there is no big biological variation between your tested subjects, then I guess it might not matter too much at the end, but I'd be cautious about the results.

Sora Yoon · 04-16-2015, 03:23 AM

Thanks for your reply. But I thought that DESeq normalization simply adjusts each column's library size for further analysis. So, I thought that I could use DESeq to technical replicates if there there is library size difference between each of the samples. Do I have some misunderstandings???

Sora Yoon · 04-16-2015, 03:35 AM

Quote:

Originally Posted by bastianwur

The normalization is meant to take the biological variations into account, IIRC.
The variance in the technical replicates will be lesser than in biological replicates, therefore the normalization will not have the right impact (less, I'd guess).

If you're sure that there is no big biological variation between your tested subjects, then I guess it might not matter too much at the end, but I'd be cautious about the results.

Thanks for your reply. But I thought that DESeq normalization simply adjusts each column's library size for further analysis. So, I thought that I could use DESeq to technical replicates if there there is library size difference between each of the samples. Do I have some misunderstandings???

fanli · 04-16-2015, 08:04 AM

Simon and others can correct me if I'm wrong, but the per-library normalization strategy taken by DESeq should be applicable to technical replicates as well. It is the model for differential expression that models biological variance rather than technical.

You should also know that technical variance in RNA-seq is generally considered trivial compared to biological..

dpryan · 04-16-2015, 10:40 AM

The normal method is to simply sum the counts across technical replicates and use those. In fact, this is also by far the most legitimate method to deal with technical replicates, unless you want to use a tool that implements mixed-effects modeling (and then you get to deal with all of the annoyances of mixed-effects models). I would be surprised if not doing this didn't lead to aberrant results, even if you try to throw extra model terms in to compensate for the design aberrantly thinking it has more samples than it really does.

Edit: BTW, as fanli wrote, the library size normalization itself shouldn't have any issues.

Caroline336 · 04-24-2015, 02:54 AM

Hi, maybe you can ask the professionals in CD Genomics, they can help you with your problem.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
DESeq without replicates	austinpa	Bioinformatics	43	07-15-2014 07:38 PM
Help Using DESeq without replicates	aprilw	Bioinformatics	6	12-19-2013 08:07 AM
No Replicates Cuffdiff; How does the 'variability model good fit' apply? and other...	Daniel1977	RNA Sequencing	6	12-02-2013 04:57 PM
attempt to apply non-function while trying DEseq vignette sample data	LanaG	Bioinformatics	7	10-25-2011 01:19 PM
DESeq: question about with replicates and without any replicates.	nb509	RNA Sequencing	2	10-25-2011 07:04 AM

04-16-2015, 01:47 AM	#1
Sora Yoon Junior Member Location: Ulsan, S.Korea Join Date: Apr 2015 Posts: 8	Can I apply DESeq to techinical replicates?? Hi guys, I've been dealing with RNA-seq count data consisting of several technical replicates. And I tried to normalize the data using DESeq. However, the DESeq author suggested to apply DESeq to biological replicates not to technical replicates. Anyone knows the reason for this?? Why can't I apply DESeq normalization to my technical replicates???? Also, is there any method to normalize a technical replicates effectively?? Thank you for reading.

04-16-2015, 03:23 AM	#3
Sora Yoon Junior Member Location: Ulsan, S.Korea Join Date: Apr 2015 Posts: 8	a confused point... Thanks for your reply. But I thought that DESeq normalization simply adjusts each column's library size for further analysis. So, I thought that I could use DESeq to technical replicates if there there is library size difference between each of the samples. Do I have some misunderstandings???

04-16-2015, 03:03 AM	#2
bastianwur Member Location: Germany/Netherlands Join Date: Feb 2014 Posts: 98	The normalization is meant to take the biological variations into account, IIRC. The variance in the technical replicates will be lesser than in biological replicates, therefore the normalization will not have the right impact (less, I'd guess). If you're sure that there is no big biological variation between your tested subjects, then I guess it might not matter too much at the end, but I'd be cautious about the results.

04-16-2015, 08:04 AM	#5
fanli Senior Member Location: California Join Date: Jul 2014 Posts: 198	Simon and others can correct me if I'm wrong, but the per-library normalization strategy taken by DESeq should be applicable to technical replicates as well. It is the model for differential expression that models biological variance rather than technical. You should also know that technical variance in RNA-seq is generally considered trivial compared to biological..

04-16-2015, 10:40 AM	#6
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,480	The normal method is to simply sum the counts across technical replicates and use those. In fact, this is also by far the most legitimate method to deal with technical replicates, unless you want to use a tool that implements mixed-effects modeling (and then you get to deal with all of the annoyances of mixed-effects models). I would be surprised if not doing this didn't lead to aberrant results, even if you try to throw extra model terms in to compensate for the design aberrantly thinking it has more samples than it really does. Edit: BTW, as fanli wrote, the library size normalization itself shouldn't have any issues.

04-24-2015, 02:54 AM	#7
Caroline336 Junior Member Location: Shirley Join Date: Apr 2015 Posts: 4	Hi, maybe you can ask the professionals in CD Genomics, they can help you with your problem.