DESeq factor design, and optimal number of samples

Lugalbanda · 10-05-2016, 02:49 PM

For the 1st question, I want to know if I should run a single factor or double factor. These samples are match normal/tumor from different patients so something like this

p1N p1T p2N p2T p3N p3T

do i have to run a double factor run like this:

------------condition---------patientNumber
------------<factor>--------- <factor>
p1N--------normal------------ 1
p1T----------tumor ------------1
p2N---------normal-------------2
p2T----------tumor--------------2
p3N--------normal-------------3
p3T---------tumor--------------3

or a single factor like this:

-------------Condition
-------------<factor>
p1N--------normal
p1T---------tumor
p2N--------normal
p2T---------tumor
p3N--------normal
p3T---------tumor

I suppose the question boils down to whether or not DESeq cares if samples come from different patients.

For the 2nd question, I have a thousand match normal/tumor from the same type of tissue but from different patients, should I run DESeq on all one thousand of them or is it fine to do a smaller amount like one hundred?

dpryan · 10-06-2016, 12:35 AM

If you have tumour/normal pairs, then you should include the pairing information

Code:

design = ~patientNumber + condition

DESeq2 (there's no reason to use DESeq) can get rather slow with 1000 samples. You'd be better with limma in that case.

SylvainL · 10-06-2016, 12:48 AM

Hi dpryan,

how slow for 1000 samples?

dpryan · 10-06-2016, 12:52 AM

@SylvainL: Good question and I don't know the answer to that. Perhaps Mike Love will pop by the thread and give a guesstimate.

fanli · 10-06-2016, 01:56 PM

It seems from the cancer literature that most/all studies using a paired tumor-normal design only have one sample from each (e.g. p1T, p1N) and not in replicate (e.g. p1Ta, p1Tb, p1Tc, p1Na, p1Nb, p1Nc). Is that appropriate, or would it be better to do things in triplicate? Say by splitting a tumor sample into 3 sections?

SylvainL · 10-07-2016, 01:50 AM

Your replicates are paired (p1T, p1N, p2T, p2N, etc...) so there is no need to split.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Optimal read length and number for plant virus discovery	HFSG	Illumina/Solexa	0	08-21-2015 10:39 AM
Design for normalization using DESeq	Ayaka	Bioinformatics	0	03-07-2015 05:21 PM
DESeq, experimental design	lmolokin	Bioinformatics	14	06-12-2013 07:36 AM
DESeq Design Issue	jpollack	Bioinformatics	0	02-04-2013 12:44 PM

10-05-2016, 02:49 PM	#1
Lugalbanda Junior Member Location: Canada Join Date: Oct 2016 Posts: 3	DESeq factor design, and optimal number of samples For the 1st question, I want to know if I should run a single factor or double factor. These samples are match normal/tumor from different patients so something like this p1N p1T p2N p2T p3N p3T do i have to run a double factor run like this: ------------condition---------patientNumber ------------<factor>--------- <factor> p1N--------normal------------ 1 p1T----------tumor ------------1 p2N---------normal-------------2 p2T----------tumor--------------2 p3N--------normal-------------3 p3T---------tumor--------------3 or a single factor like this: -------------Condition -------------<factor> p1N--------normal p1T---------tumor p2N--------normal p2T---------tumor p3N--------normal p3T---------tumor I suppose the question boils down to whether or not DESeq cares if samples come from different patients. For the 2nd question, I have a thousand match normal/tumor from the same type of tissue but from different patients, should I run DESeq on all one thousand of them or is it fine to do a smaller amount like one hundred? Last edited by Lugalbanda; 10-05-2016 at 02:53 PM. Reason: formatting

10-06-2016, 12:35 AM	#2
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,479	If you have tumour/normal pairs, then you should include the pairing information Code: design = ~patientNumber + condition DESeq2 (there's no reason to use DESeq) can get rather slow with 1000 samples. You'd be better with limma in that case.

10-06-2016, 12:48 AM	#3
SylvainL Senior Member Location: Geneva Join Date: Feb 2012 Posts: 175	Hi dpryan, how slow for 1000 samples?

10-06-2016, 12:52 AM	#4
dpryan Devon Ryan Location: Freiburg, Germany Join Date: Jul 2011 Posts: 3,479	@SylvainL: Good question and I don't know the answer to that. Perhaps Mike Love will pop by the thread and give a guesstimate.

10-06-2016, 01:56 PM	#5
fanli Senior Member Location: California Join Date: Jul 2014 Posts: 198	It seems from the cancer literature that most/all studies using a paired tumor-normal design only have one sample from each (e.g. p1T, p1N) and not in replicate (e.g. p1Ta, p1Tb, p1Tc, p1Na, p1Nb, p1Nc). Is that appropriate, or would it be better to do things in triplicate? Say by splitting a tumor sample into 3 sections?

10-07-2016, 01:50 AM	#6
SylvainL Senior Member Location: Geneva Join Date: Feb 2012 Posts: 175	Your replicates are paired (p1T, p1N, p2T, p2N, etc...) so there is no need to split.