Error with Cufflinks, ran out of memory

[ykingh]

My job ran out of all availabel memory when I used Cufflinks. My input file is the output from Tophat. Does anyone know how to split my input into smaller chunks or in some other way to reduce the the amount of memory my program allocates?

[zorph]

Did you ever fun out Whether you could do this?
I have the Same problem and it would be a lot faster to run cufflinks on each chromosome independtly but I'm afraid of what that will do to my fpkm values.

Has anyone done this?

[kjlee]

Hey y'all,

I am having a similar problem: cufflinks on a 10BG+ bam file leads to huge memory usage (and subsequent job stop). zorph, your idea of runnign cufflinks on each chromosome separately sounds good. I wouldn't worry about it messing up your fpkm because it is probably wise not to use fpkm to quantify expression. Instead, use a normalization scheme based on edgeR or DESeq methods. Both programs come up with library size estimates that are robust against a few differentially expressed genes unlike fpkm.
So the flow would be:
1. split cufflinks runs by chromosome
2. use the output from cufflinks (transcripts.gtf) as your gene models, and then run an overlapping program (like bedops; http://code.google.com/p/bedops/) to get the RNA-seq counts for each gene
3. normalize the counts using DESeq or edgeR
4. do your statistical comparison (DESeq is good for this step, too; here is a really straight-forward vignette: http://bioconductor.org/packages/dev.../doc/DESeq.pdf)

cheers,