GPU-enabled DeepVariant via Singularity, not Docker

deepvariant.sh

module load singularity

# we have to transform the docker image into a singularity thingy. Be careful to use the GPU docker image 0.9.0-gpu, not 0.9.0
# the following command doesn’t work on copyQ?? I get `mksquashfs not found`, works fine on other systems

singularity pull $MYGROUP/singularity/myRepository/deepvariant.sif docker://google/deepvariant:0.9.0-gpu
# you may have to make the above folders

Then, using the example data described here: https://github.com/google/deepvariant/blob/r0.9/docs/deepvariant-quick-start.md
INPUT_DIR="${PWD}/quickstart-testdata" 
DATA_HTTP_DIR="https://storage.googleapis.com/deepvariant/quickstart-testdata"

mkdir -p ${INPUT_DIR}
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/NA12878_S1.chr20.10_10p1mb.bam
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/NA12878_S1.chr20.10_10p1mb.bam.bai
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/test_nist.b37_chr20_100kbp_at_10mb.bed
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/test_nist.b37_chr20_100kbp_at_10mb.vcf.gz
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/test_nist.b37_chr20_100kbp_at_10mb.vcf.gz.tbi
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/ucsc.hg19.chr20.unittest.fasta
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/ucsc.hg19.chr20.unittest.fasta.fai
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/ucsc.hg19.chr20.unittest.fasta.gz
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/ucsc.hg19.chr20.unittest.fasta.gz.fai
wget -P ${INPUT_DIR} "${DATA_HTTP_DIR}"/ucsc.hg19.chr20.unittest.fasta.gz.gzi

OUTPUT_DIR="${PWD}/quickstart-output"
mkdir -p "${OUTPUT_DIR}"

# Then, on Topaz within a gpu-dev or gpu queue
salloc -p gpuq-dev --gres=gpu:1
# alternatively within a proper sbatch script of course

# run the actual command
export containerImage=$MYGROUP/singularity/myRepository/deepvariant.sif

srun --export=ALL singularity run --nv $containerImage    /opt/deepvariant/bin/run_deepvariant   --model_type=WGS --ref={INPUT_DIR}/ucsc.hg19.chr20.unittest.fasta   --reads=${INPUT_DIR}/NA12878_S1.chr20.10_10p1mb.bam   --regions "chr20:10,000,000-10,010,000"   --output_vcf=${OUTPUT_DIR}/output.vcf.gz   --output_gvcf=${OUTPUT_DIR}/output.g.vcf.gz   --num_shards=1

# the --nv flag is very important. without it the NVIDIA support isn't there and it complains about missing libcuda
# it should only run for 5 seconds or so