GATK 3.8 was released on July 28, 2017. Itemized changes are listed below. For more information on what you should care about, see the user-friendly version highlights (coming soon).
Performance improvements / accelerations
- Add the Intel GKL as a dependency, and turn on the Intel inflater/deflater by default
- Use GKL's implementation of Vector Logless Pair HMM (HaplotypeCaller and MuTect2)
- Add FPGA support for PairHMM via GKL (experimental)
Notable tool behavior change
- In VariantsToTable, flip handling of missing annotation values: allow by default (print NA for null object), deprecate --allowMissingData, add --errorIfMissingData flag
Improvements to germline calling toolchain
- Change calculation of rank sum to be the median of the per-sample rank sums (because the rank sum across all sample data did worse for biallelics)
- Improve VQSR model serialization functionality and allow serialized model reports to be read into the variant recalibrator to recreate a GMM from a previous VQSR run
- Allow disabling of BadMateFilter in HaplotypeCaller
Bug fixes
- Fix LeftAlignAndTrimVariants -split to not change no-call genotypes to hom-ref
- Fix ClippingOP.getNewAlignmentStartOffset() to handle N cases properly
- Fix FindCoveredIntervals so -minBQ, -minMQ and -cov work as expected
- Fix thread safety bugs (upstreamDeletionsLoc and practicalAlleleCountForPloidy)
- Fix bug in depth of coverage for partitioning by library
- In GenotypeGVCFs, total ploidy does not have to equal the number of priors
- Explicitly report AD and DP as zero if there is no read coverage
- GenotypeGVCFs and CombineGVCFs now throw errors when AS annotations are not present if allele-specific mode is activated
- Copy Program Records (PG) from input BAM to -bamout file (failure to do so caused some validation failures)
Docs / logging
- Assorted corrections and improvements in documentation, error messages and logging rules
- Archive GATK3-specific docs from the forum
Misc engine
- Bump htsjdk to ver 2.10.1 and Picard to ver 2.10.2