Parabricks

This article is a draft

This is not a complete article: This is a draft, a work in progress that is intended to be published into an article, which may or may not be ready for inclusion in the main wiki. It should not necessarily be considered factual or authoritative.

Parabricks is a software suite for performing secondary analysis of next generation sequencing (NGS) DNA data. A major benefit of Parabricks is that it is designed to deliver results at blazing fast speeds and low cost. Parabricks can analyze whole human genomes in about 45 minutes, compared to about 30 hours for 30x WGS data. The best part is the output results exactly match the commonly used software. So, it's fairly simple to verify the accuracy of the ouput.

Under the hood, it achieves this performance through tight integration with GPUs, which excel at performing data parallel computation much more effectively than traditional CPU-based solutions. Parabricks was built from the ground up by GPU computing and Deep Learning experts who wanted to develop the fastest and most efficient possible implementation of common genomics algorithms used in secondary analysis.

You can learn more at www.nvidia.com/parabricks

Usage in Compute Canada Clusters

This software has been provided freely by NVidia to help with research on COVID19 until Sunday, 17 May 2020

Finding and loading Parabricks

Parabricks can be looked for as a regular module through module spider:

[name@server ~]$ module spider parabricks

Likewise, it can be loaded through LMOD modules:

[name@server ~]$ module load parabricks/2.5.0

Example of usage

Before you embark on using Parabricks, make sure you have gone through the Parabricks documentation, including their standalone tools and pipelines. Also make sure you know how to request graphic cards in Compute Canada clusters. Once you understand the above, you can submit a job like:

#!/bin/bash
#SBATCH --account=def-someuser
#SBATCH --gres=gpu:1
#SBATCH --nodes=1
#SBATCH --cpus-per-task=32
#SBATCH --mem=0
#SBATCH --time=5:00:00

module load parabricks/2.5.0

DATA_DIR=/path/to/data
OUT_DIR=/path/to/output
pbrun germline \
      --ref ${DATA_DIR}/Homo_sapiens_assembly38.fa \
      --in-fq ${DATA_DIR}/some_1.fastq ${DATA_DIR}/some_2.fastq \
      --knownSites ${DATA_DIR}/dbsnp_146.hg38.vcf.gz \
      --tmp-dir ${SLURM_TMPDIR} \
      --out-bam ${OUT_DIR}/output.bam \
      --out-variants ${OUT_DIR}/output.vcf \
      --out-recal-file ${OUT_DIR}/report.txt

Make the path to the files absolute real paths (i.e. with the command realpath .)!!

Common issues

Almost immediate fail

If your first test fails right away, there might be a missing module or some environmental variable clash. To solve this you can:

[name@server ~]$ module --force purge

[name@server ~]$ module load StdEnv/2016.4 nixpkgs/16.09 parabricks/2.5.0

And you should be ready to go.

Later fail

Oftentimes Parabricks would not give you a clear traceback of the fail. This usually means that that you did not requested enough memory. If you are reserving a full node already through --nodes=1, we suggest you also use all the memory in the node with --mem=0. Otherwise, make sure that your pipeline has enough memory to process your data.

Hybrid usage

Parabricks uses both CPU and GPUs. During our tests, Parabricks used at least 10 CPUs, so we recommend to ask for at least that amount through --cpus-per-task=10

References

Parabricks Home