how to speed up blastx run on slurm HPC

Question

[Bioinfonext]

Hi,

I am using below script for balstx against the ncbi nr database but it is taking so much time, and I got around 150000 sequences in fasta file? Is there any way to speed up the analysis?

#!/bin/bash
#SBATCH --job-name=nr      # Job name
#SBATCH --mail-type=END,FAIL         # Mail events (NONE, BEGIN, END, FAIL, ALL) 
#SBATCH --ntasks=40                  # Number of MPI ranks
#SBATCH --nodes=2                    # Number of nodes
#SBATCH --time=8000:00:00              # Time limit hrs:min:sec
#SBATCH --partition=k2-lowpri
#SBATCH --mem=200G

module load apps/ncbiblast/2.10.0/gcc-7.2.0

blastx -query sequecnes.fasta -db /mnt/scratch/ncbi/nr_protein/nr -evalue 1e-5 -outfmt "7 std stitle qseqid sseqid pident length mismatch gapopen qstart qend qcovs qlen slen" -max_target_seqs 1 -out nr.new.txt

Many thanks

Answer 1

[GenoMax]

By using a corresponding request in your blastx command line of -num_threads to match --ntasks. I would keep the threads on one node so change --nodes to 1.

[Bioinfonext]

thanks genomax,

so should I change the resources like this:

#!/bin/bash
#SBATCH --job-name=nr      # Job name
#SBATCH --mail-type=END,FAIL         # Mail events (NONE, BEGIN, END, FAIL, ALL) 
#SBATCH --ntasks=40                  # Number of MPI ranks
#SBATCH --nodes=1                    # Number of nodes
#SBATCH --time=8000:00:00              # Time limit hrs:min:sec
#SBATCH --partition=k2-lowpri
#SBATCH --mem=200G

many thanks

[GenoMax]

Correct. As long as individual nodes have 40 cores. If not drop that number to match what your nodes have. You can also used DIAMOND (LINK) as a faster alternative. You will need to create your own indexes though.

Answer 2

[Mensur Dlakic]

Add -num_threads 40 to your blastx command.

[Bioinfonext]

thank you so much! now I understand it.