GATK release 1.4

Release date: December 31, 2011

Unified Genotyper

• Implemented the proper QUAL calculation for multi-allelic calls (used previously only in GENOTYPE_GIVEN_ALLELES mode).
• The ability to genotype multiple alleles in discovery mode has been added for SNPs. See the --multiallelic argument documentation for more details. Please note that this is still highly experimental.
• Added an argument to specify the maximum number of alternate alleles to genotype at any one site.
• When making reference calls, don't output per-sample PLs (since they don't make sense without an alternate allele).
• Genotypes are now assigned with a greedy approach to achieve the target MLE, instead of a dynamic programming implementation based on the posterior probabilities.
• No-call genotypes are no longer annotated with garbage values.
• Bug fix for indels: skip likelihood computation in the case where too many bases were cut and there is no haplotype or read left.
• Bug fix for indels: catch out of bound error that can show up in RNAseq reads with long N CIGAR operators in the middle.

Variant Annotator

• Added an annotation to compute the TDT (Transmission Disequilibrium Test).
• Added an option to do strict allele matching when annotating with comp track concordance.
• QualByDepth now correctly skips no-call genotypes.
• All Rank Sum Tests were updated to be deterministic.
• Transfers headers from the resource VCF (when possible) when using expressions.
• Temporarily disabled SnpEff support in the GATK until SnpEff 2.0.4 has been officially released.

Variant Eval

• Better Mendelian Violation evaluation module (completely rewritten).
• Added a IntervalStratification module.
• Ti/Tv calculation no longer rounds values.
• Prints 0.0 Ti/Tv and not NaN when there are no variants.
• Performance optimizations throughout the tool.
• Tables now are cleanly formatted (floats are %.2f printed).
• VariantSummary is a standard report now.
• Removed CompEvalGenotypes (it didn't do anything) and SimpleMetricsByAC modules.

Select Variants

• Fixed bug that occurred when multiple records were present at a given position, where all variants that follow after one that fails filters are dropped (instead of dropping just the failing one).
• Added support for selecting only variants with specific IDs from a file.
• Added support for running multi-threaded (i.e. with the -nt argument).
• Fixed incorrect performance when using a discordance track but no sample specifications.

Combine Variants

• Implemented a better algorithm for merging genotypes.

Indel Realigner and Realigner Target Creator

• No longer changes the mapping quality of MQ=255 reads that get realigned.
• Fixed bug where the GATK holds onto too much memory when running the Realigner Target Creator with multiple threads.

Variant Quality Score Recalibrator

• Now detects if the negative model failed to converge properly because of having too few data points and automatically retries with more appropriate clustering parameters.

Variant Context

• Fixed bug for complex records: we can no longer clip out a complete allele.
• We now fail gracefully when encountering malformed VCFs without enough data columns.

Misc Tools and Features

• We now check for reads with missing read groups and throw a UserException when encountered.
• Rarely (if ever) used Read Pair traversal type is no longer supported.
• We no longer allow the '-L "interval1;interval2"' syntax; use '-L interval1 -L interval2' instead.
• Fixed a bug in the automatic clipping of adapter sequence in LocusWalkers.
• Fixed a performance issue with RODWalker parallelization.
• Performance improvements to the general GATK engine, especially when working with many input BAM files.
• Added a new tool called Phase By Transmission to perform phasing of variants based on family structure.
• Added a new tool called Validation Site Selector that helps the user choose a subset of sites to use for a validation of the overall callset.
• Fix for problem with too many open file handles: actually close file handles when close() is called.
• Fix for problem when using Bed files as intervals: one no longer needs to explicitly state the file type.
• Traversal statistics start printing after 30 seconds instead of 2 minutes.
• Revved Picard jar to version 1.57.1030.
• Revved Tribble jar to version 46.

This includes up to commit 472fc94f003d7043614639dfde29b38cb7a9d439.

For GATK developers: new VariantContext interface

GenotypesContext: an efficient representation for a collection of Genotype objects

• Old version was a Map<String, Genotype>. New version is a GentotypesContext. Implements List<Genotype> interface, but also has efficient random access (getSample(String sampleName), for example)

• Static constructor interface (GenotypesContext.create() and GenotypesContext.copy()) is used to create:

// create an empty GC
GenotypesContext gc1 = GenotypesContext.create();

// create an GC containing a copy of the GenotypesContext from the VariantContext
GenotypesContext gc2 = GenotypesContext.copy(variantContext.getGenotypes());

VariantContexts can now only be built with VariantContextBuilder

• Old approach had many overloaded constructors. Lots of code duplication of creating empty data structures for missing arguments. What we really need is a clean programming approach to creating a VariantContext. VariantContextBuilder follows Java builder design to easily create VariantContexts:

// create an empty builder with all default values
VariantContextBuilder def = new VariantContextBuilder();

// create a VCB with the min. amount of information to make a VariantContext, and set the log10PError
VariantContextBuilder builder = new VariantContextBuilder("name", loc.getContig(), loc.getStart(), loc.getStop(), alleles).log10PError(0.0);

// create a new variant context builder based on the old one
VariantContextBuilder builder = new VariantContextBuilder(vc);

// create a new variant context based on the old one
VariantContext copy = new VariantContextBuilder(vc).make();

// create a new variant context based on the old one with new genotypes
VariantContextBuilder sub = new VariantContextBuilder(vc).genotypes(genotypes);

A non-exhaustive list of other important changes

• The annoying representation of probabilities as -1 * log10(p) is gone. All code uses log10(p) instead
• Massive UnitTests everywhere are ensuring correctness. Many edge-case bugs are fixed
• Underlying GenotypesContext and associated cleanups have speed up the code significantly. For example, working with HM3 and OMNI genotypes VCFs, we see:
  • CombineVariants is 33% faster with the new code
  • SelectVariants for NA12878 is 33% faster
  • VariantEval with -ST Sample on theOMNI file is 33% faster
• isPolymorphic is now isPolymorphicInSamples (same for isMonomorphic)
• getChromosomeCount is now getCalledChrCount to reflect that is now treats genotype A/. as 1 called chromosome
• subContextFromGenotypes completed removed in favor of cleaner subContextFromSamples
• ID field is a first class object in VariantContext -- no longer stored in the attributes
• MutableVariantContext and MutableGenotype removed as they never were really used.