Cufflinks memory usage

[oscarluoinau]

Hi all,
I am having trouble with running Cufflinks on PE RNA-Seq libraries generated from HiSeq machine. I have used Tophat to successfully mapped those PE reads (about 160 millions reads), which gave me a BAM file about 6GB for each library. Then I fed the BAM file to Cufflinks running with 20 cores. Now the problem is it seems Cufflinks is taking more than 120GB ram, and is taking very long (about a week) to run one library. Have any of you had similar experience? Am I doing something wrong? Any suggestions? Thanks!

[Nicolas]

Which reference annotations are you using? I had similar experience with gencode annotations (>2.5 millions annotations). I then switched to RefSeq annotations and the Cufflinks runs are now much shorter. Of course, it's quantify much less potential transcripts, but for most applications, that can be sufficient.

[jbrwn]

in addition to posting your reference, you may want to post which options you're utilizing in cufflinks.

[oscarluoinau]

Right. I am also using the Gencode annotation. I think I will experiment with other annotation file to see how it goes. The options I am utilizing in Cufflinks are simply specifying the reads come from PE reads (i.e. --fr-unstranded).

[kjlee]

oscar,

did you ever get an answer or a work around? I am running cufflinks on a similar sized bam file and am also running out of memory.

cheers,

[oscarluoinau]

I couldn't get the job done until upgraded to the latest version of Cufflinks which seems to use less memory. Good luck!

[kjlee]

Hey Oscar,

I am running the newest version of Cufflinks (I literally downloaded it this week). I got Cufflinks to run on a small sub-set of reads (about 4 million paired end reads, 100nt with ~200nt inner gap). But the whole data-set is ~50X bigger. How many reads did you use? And what (approximately) was the memory usage for your file size (RAM per GB of the bam/sam file)?

Cheers,

[oscarluoinau]

Hi,
I don't remember the exact numbers as it was about a year ago. One thing I do remember is I was lucky enough to utilize a machine with 1TB RAM, and I used about 500GB for about 160 million reads. I hope this helps. Good luck!

[kjlee]

We only have one compute node with that much memory on our cluster and I didn't want to usurp it if I didn't have to. But I guess that's what the resources are for. Thanks Oscar. (Also, about how long did it take to run?)

Cheers,

[oscarluoinau]

Expect it to run longer than a week.