About ParabricksRelease Notes

4.3.1-1 Release Notes

View as Markdown

Highlights:

  • New tool, deepsomatic for somatic variant calling.
  • The latest Parabricks toolkit (v4.3.1) is now fully supported on Grace Hopper, refer to NVIDIA Grace CPU Platforms.
  • deepvariant version 1.6.1 updated, new option added.
  • haplotypecaller bug fixes and 3 new options added.
  • mutectcaller a few bug fixes.
  • rna_fq2bam bug fixes and feature added in wrapper.
  • minimap2 bug fixes, version update, and improved support for ONT data.
  • fq2bamfast and fq2bam: Bug fixes and performance improvements. Additional option to monitor approximate CPU utilization and host memory usage during execution (--monitor-usage).

New Tools

With Parabricks 4.3.1 we are releasing an accelerated deepsomatic tool. DeepSomatic builds on the deep learning-based variant caller DeepVariant. It processes aligned reads from tumor and normal samples (in BAM or CRAM format), generates pileup image tensors, classifies these tensors using a convolutional neural network, and outputs somatic variants in standard VCF or gVCF files.

Improvements

Tool Updates

haplotypecaller:

  • Adds the following new options:

    • --minimum-mapping-quality
    • --mapping-quality-threshold-for-genotyping
    • --enable-dynamic-read-disqualification-for-genotyping

  • Improved performance by leveraging AVX512 instructions for CPU-based PairHMM computation.

mutectcaller:

  • Improved performance by leveraging AVX512 instructions for CPU-based PairHMM computation.

deepvariant:

  • Updates to match the baseline version v1.6.1.
  • Adds the new option --haploid-contigs
  • Improved performance for short-read mode through increased GPU utilization and kernel optimizations.

rna_fq2bam:

  • Supports passing --out-chim-type multiple times.

fq2bamfast

  • Improved alignment performance on Hopper GPUs through increased use of DPX instructions.
  • Improved performance on multi-GPU runs; for example, on DGX H100.
  • Improved error detection for improper FASTQ inputs through --in-fq or --in-se-fq. Previously recorded a utf-8 decode error.
  • Additional option to monitor approximate CPU utilization and host memory usage during execution (--monitor-usage).

fq2bam:

  • Improved error detection for improper FASTQ inputs through --in-fq or --in-se-fq. Previously recorded a utf-8 decode error.
  • Additional option to monitor approximate CPU utilization and host memory usage during execution (--monitor-usage).

minimap2:

  • Updated map-pbmm2 preset to match the updated versions of minimap2 (v2.26) and pbmm2 (v1.13.0).

Bug Fixes

  • mutectcaller and haplotypecaller: Fixed a wrong alignment offset value in smith-waterman algorithm.
  • mutectcaller and haplotypecaller: Fixed a crash on GPU when running in low memory mode.
  • mutectcaller: Fixed the wrong active probability value when the pileup size is 0.
  • mutectcaller: Fixed a max coverage overflow bug.
  • rna_fq2bam: Fixed an error when passing “WithinBAM_SoftClip” or “WithinBAM_HardClip” to --out-chim-type.
  • minimap2: Fixed support for Oxford Nanopore Technologies (ONT) data with minimap2.
  • fq2bamfast : Fix rare erroneous assertion (Workspace not big enough, expected desiredSize <= cubWorkspaceSize, exiting). Case will now be handled correctly and fall back to CPU recovery if needed.
  • deepvariant: Fixed bug related to Smith-Waterman computation on CPU.

For further information refer to the Parabricks datasheet.