Welcome to NVIDIA Parabricks v4.3.2
NVIDIA Parabricks v4.3.2

4.3.2-1 Release Notes

Highlights:

  • New pipeline, ont_germline for long read ONT variant calling.

  • minimap2 bug fixes and new features.

  • Further optimizations and improvements for the Grace Hopper architecture.

  • Introduced GPU acceleration for CRAM file write operations.

With Parabricks 4.3.2 we are releasing an accelerated ont_germline pipeline. The ONT Germline pipeline generates BAM and variants on long read ONT sequences using minimap2 for alignment as well as the DeepVariant variant caller.

Tool Updates

minimap2:

  • Added new options --pbmm2 and --pbmm2-unmapped that include additional processing to match the output of PacBio's pbmm2.

  • Added ability to use unaligned BAM as input instead of FASTQ (--in-bam).

  • A new higher compression ratio option for BAM/CRAM files with --gpuwrite.

  • The nvCOMP DEFLATE compression algorithm options have changed for the parameter --gpuwrite-deflate-algo. Please see --help or the documentation on fq2bam for more details.

  • Added GPU acceleration for CRAM write operations (--gpuwrite).

fq2bam:

  • fq2bam has been surpassed by fq2bamfast: the new version of fq2bam with more optimizations for performance. Note that fq2bamfast is no longer present as the fq2bam and fq2bamfast tools have been merged.

  • Added a new option to use a single-ended unaligned BAM as input instead of a FASTQ file (--in-se-bam).

  • A new higher compression ratio option for BAM/CRAM files with --gpuwrite.

  • The nvCOMP DEFLATE compression algorithm options have changed for the parameter --gpuwrite-deflate-algo. Please see --help or the documentation on fq2bam for more details.

  • Added GPU acceleration for CRAM write operations (--gpuwrite).

  • Default --min-read-length is 1 base pair; whereas, in fq2bamfast it was 10.

  • Improved error messaging if paired reads have different names.

deepvariant:

  • Added a new option --filter-reads-too-long to filter out reads with length > 512bp.

  • Improved performance and reduced memory consumption when processing long reads (PacBio and ONT data)

haplotypecaller added support for:

  • Force-calling mode (--htvc-alleles).

  • The --min-base-quality-score argument.

  • The --adaptive-pruning argument.

  • A new annotation AssemblyComplexity can be specified by passing --haplotypecaller-options '-A AssemblyComplexity'.

somatic:

  • Added more sanity checks to handle possible errors in advance.

germline pipeline and deepvariant_germline:

  • Grace Hopper (GH200)-specific optimizations.

fq2bam_meth:

  • Default --min-read-length is now 1 base pair to be in line with fq2bam.

  • Improved error messaging if paired reads have different names.

minimap2:

  • Fixed a memory leak that was exposed by using a different matching score in the updated preset values.

  • Fixed a bug that caused some incorrect SA tag values.

  • Fixed a crash caused by pthread_setaffinity_np.

fq2bam (BWA-MEM + GATK):

  • Fixed a crash when passing --bwa-options="-C" with no comments in the input fastq file(s).

mutectcaller:

  • Fixed an invalid GPU memory access in force calling mode (--mutect-alleles).

  • Fixed additional records in output VCF file when passing --pon.

somatic (Somatic Variant Caller):

  • Fixed a read group ID conflict error in tumor-normal mode when --tumor-read-group-id-prefix was not passed from the user.

haplotypecaller:

  • Fixed an out-of-order issue in the output BAM when passing --htvc-bam-output.

deepvariant:

  • Fixed a bug where certain kernels from multiple pipelines (streams) were launched on the default stream.

For further information see the Parabricks datasheet.

Previous What's New?
Next 4.3.1-1 Release Notes
© Copyright 2024, Nvidia. Last updated on Sep 17, 2024.