broadinstitute/gatk 4.0.8.0

on GitHub

This release features some significant changes to Mutect2 that improve both performance and correctness, as well as a bug fix to GenomicsDBImport for large interval lists.

As usual, a docker image for this release can be downloaded from https://hub.docker.com/r/broadinstitute/gatk/

Full list of changes in this release:

• Mutect2

  • Handle overlapping mates in M2 active region detection, causing fewer false active regions (#5078)
    • Makes Mutect2 ~25% faster in many cases with no loss of accuracy!
  • Filter M2 calls that are near other filtered calls on the same haplotype (#5092)
    • A very effective new filter that significantly reduces false positives
  • New Orientation Bias Filter (#4895)
    • New, improved orientation bias model, without which the M2 pipeline is not viable for NovaSeq data.
  • Changed the default AF slightly for M2 tumor-only mode (just a small tweak) (#5067)
  • Optimize some Mutect-related tools (#5073)
    • Everything that inherits from AbstractConcordanceWalker (this includes the Concordance tool and MergeMutect2CallsWithMC3) is now much faster on the cloud
  • Fixed edge case for M2 palindrome transformer (#5080)
    • Fixed an edge case involving reads assigned huge fragment lengths
  • Allowing counts for supporting alt reads in the validation normal. (#5062)
    • Added useful information suggesting possible normal artifacts in somatic validation tool.
  • M2 wdl doesn't emit unfiltered vcf, which is redundant (#5076)

• GenomicsDBImport

  • Fix for issue where we could run out of file handles when working with large interval lists (#5105)
  • Display warning when using large interval lists with GenomicsDBImport (#5102)

• Updated MarkDuplicatesSpark tie-breaking rules to reflect changes in picard (#5011)

• Added the ability for CompareDuplicatesSpark to output mismatching reads (#4894)

• Updated our google-cloud-java fork to 0.20.5-alpha-GCS-RETRY-FIX (#5099)

  • We now retry on 502 and UnknownHostException errors when using NIO

• SV Tools:

  • Various improvements (#4996)
    • output a single VCF for new interpretation tool
    • bring MAX_ALIGN_LENGTH and MAPPING_QUALITIES annotations from CPX variants to re-interpreted simple variants
    • add new CLI argument and filter assembly based variants based on annotation MAPPING_QUALITIES, MAX_ALIGN_LENGTH
    • filter out variants of size < 50
  • Bug fix for the extreme edge case where after alignments de-overlapping, an alignment block is only 1 base long (#4962)
  • Turn back on checking variant info fields against header in SV vcf writing (turned off temporarily long time ago but slipped attention after implementation stablized) (#5084)