A problem with Elastic net + Cox regression

Anastaziel · 02-11-2016, 03:14 PM

I was trying to perform elastic net + cox regression in R using the glmnet package. The problem is that my data is around 100 individuals with ~100k features and the model is too big for R to store in a matrix. Although the glmnet package can use sparse matrix for the model unfortunately it does not support it with Cox regression.

Anyone knows of another tool I can use (I am not locked in R, it was just the only tool I found)? I read that I can use an SVM (support vector machine) but I still need to calculate the model. I am looking for any tool (R, Matlab, stand alone application) that can perform elastic net + cox regression on a big model. Furthermore, maybe you know how to calculate this big model as a stand alone output so I can use it in an SVM?

GenoMax · 02-11-2016, 03:45 PM

For reference cross-posted: https://www.biostars.org/p/176689/

Anastaziel · 02-11-2016, 03:50 PM

I am sorry, I did not know that it is against the rules to post in multiple forums the same question. Not to make another post on this: I will mention if on the other forum the question in answered.

GenoMax · 02-11-2016, 03:54 PM

Quote:

Originally Posted by Anastaziel

I am sorry, I did not know that it is against the rules to post in multiple forums the same question.

No it is not against the rules so don't worry.

I included the link so people can see if your question has been answered on Biostars before they compose a lengthy response (if one was needed) here.

You can help by coming back and posting a closure if your question is answered on Biostars.

Anastaziel · 02-12-2016, 01:53 AM

I was able to fix my problem. I was using a wrong test case and my model was becoming too big. Using the example from glment.pdf fixes the issue.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
FPKM is 0, when using cufflinks in galaxy on the net	chenyao	Bioinformatics	1	12-04-2012 07:12 AM
.NET Bio 1.0 training course	rickbe	Events / Conferences	0	10-28-2011 09:28 AM
BFAST to Sourceforge.net	nilshomer	Bioinformatics	9	03-31-2011 10:28 AM

02-11-2016, 03:14 PM	#1
Anastaziel Junior Member Location: Moon Join Date: Jun 2015 Posts: 7	A problem with Elastic net + Cox regression I was trying to perform elastic net + cox regression in R using the glmnet package. The problem is that my data is around 100 individuals with ~100k features and the model is too big for R to store in a matrix. Although the glmnet package can use sparse matrix for the model unfortunately it does not support it with Cox regression. Anyone knows of another tool I can use (I am not locked in R, it was just the only tool I found)? I read that I can use an SVM (support vector machine) but I still need to calculate the model. I am looking for any tool (R, Matlab, stand alone application) that can perform elastic net + cox regression on a big model. Furthermore, maybe you know how to calculate this big model as a stand alone output so I can use it in an SVM?

02-11-2016, 03:50 PM	#3
Anastaziel Junior Member Location: Moon Join Date: Jun 2015 Posts: 7	I am sorry, I did not know that it is against the rules to post in multiple forums the same question. Not to make another post on this: I will mention if on the other forum the question in answered. Last edited by Anastaziel; 02-11-2016 at 04:13 PM.

02-11-2016, 03:45 PM	#2
GenoMax Senior Member Location: East Coast USA Join Date: Feb 2008 Posts: 7,087	For reference cross-posted: https://www.biostars.org/p/176689/

02-12-2016, 01:53 AM	#5
Anastaziel Junior Member Location: Moon Join Date: Jun 2015 Posts: 7	I was able to fix my problem. I was using a wrong test case and my model was becoming too big. Using the example from glment.pdf fixes the issue.