STAR Aligner

[sbdk82]

I am running STAR for aligning wheat RNA-Seq data with Ensemble reference file . The size of reference file is 4gb. The genome directory created in the first step is 42 gb. The mapping step took more than 50 hours. Some jobs are still running for more than 75 hours

I used 5 nodes with 100gb each in our university cluster . Here is the script I used

#!/bin/sh
#SBATCH --job-name=STAR
#SBATCH --nodes=5
#SBATCH --ntasks-per-node=1
#SBATCH --time=120:00:00
#SBATCH --mem=100g
#SBATCH --error=<Error File Name>
#SBATCH --output=<Output File Name>

cd  /Dir_PATH/STAR

./STAR_2.4.0b/STAR --genomeDir /Dir_PATH/STAR/index  --readFilesIn  /File_PATH/L001_R1_001.fastq,/File_PATH/L002_R1_001.fastq File_PATH/L001_R2_001.fastq,/File_PATH/_L002_R2_001.fastq --outFileNamePrefix /Dir_PATH/<Prefix_Name>/ --runThreadN 10

I ran BWA-MEM on same data and it took less than 10 hours to complete the mapping. Am I doing something wrong or do I need to choose some other parameters ?

[dpryan]

That seems quite odd. You're giving each of the nodes different files, yes?

[sbdk82]

I think it uses total 500 gb (100gb x 5 nodes) for this job. It does not distribute different files into different nodes.

[dpryan]

You're just running the same thing on all of the nodes. If you're doing the same with bwa mem then that's happening there as well.

[dpryan]

I'll add that loading the I/O overhead of loading the index and constantly overwritting itself could cause a slow down (I limit STAR two 4 concurrent instances on our cluster when outputting to SAM since otherwise I can't guarantee that the drives can keep up if any other jobs are running).

[sbdk82]

So I should try with this?

#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
#SBATCH --time=120:00:00
#SBATCH --mem=100g

[dpryan]

Sure, though you don't need to specify --ntasks-per-node when you just use one node. For reference, here is the start of mine:

#!/bin/bash
#SBATCH -J STAR-align
#SBATCH -t 4:00:00
nNodes=4
#SBATCH -N 4
#SBATCH -A ryand
#SBATCH --exclusive
#SBATCH --partition=work
BIN=$WORK/bin
i=0
for i in `seq $nNodes`
do
    j=$(($i-1))
    srun -N 1 --relative $j $BIN/slurm_STAR.sh $j $nNodes &
done
wait
rm Aligned.out.sam Log.out Log.progress.out
rm -rf _STARtmp

The slurm_STAR.sh shell script will align every Nth pair of fastq files (or single fastq file, as appropriate) in a preset directory. Every instance is run on an individual node. Note that I highly recommend using --exclusive if that's not otherwise the default on your cluster.

[sbdk82]

Thanks !!! I will try that