Sorting a very large bam file

[lre1234]

Hi all,
I am doing some whole genomes. The samples were done by sequencing 1 sample on all 12 lanes of a solid 5500xl machine. From the output, I aligned each individual lane separately, then merged the 12 bam files together using picardtools mergeSam. This merged bam file is rather large, 200GB in size and has around 2billion sequence reads. My problem is that now I have to sort the file. Any method that I have tried so far has not worked. Samtools fails with an error saying that there are too many files, and picard tools gives an error that it is out of memory. The compute node that I am using has 256GB of memory, so that should be enough. I have even tried increasing the -m flag to 16G in samtools sort. Still not working.

Any thoughts?

Thanks

[GenoMax]

Check the output of the

limit

command to see if you are running into a constraint in user resources. If you are submitting the job using a scheduler (LSF/SGE) then it would be useful to check if the queue you are using has a memory limit (per job/user)?

[Richard Finney]

Use as much memory as you can. Set it to 240 GB or so if you have 256 and nothing else is running. Make sure you are using the -m parameter properly.
The line to parse the -m arg is

case 'm': max_mem = atol(optarg); break;

There's no "16GB" , it's got to be 16000000000 (if I counted the zeros right).

[lh3]

More recent versions parse "16G". 0.1.18 definitely does not.

[maubp]

It might be worth presorting the 12 individual BAM files before merging, that should make the final sort quicker.

(The extra IO might overall be slower)

[lre1234]

So, my 12 original files were all pre-sorted before merging, but I did manage to get the file sorted.

Originally, I was using samtools 0.1.18, and with the sort command I set the -m flag to 16000000000. This took ~12 hrs to finish.

After updating to samtools 0.1.19 and using the sort command with -m 2G and -@ 32, the file sorted in about 4hrs...

Much improvement