This is to facilitate the current experiment with class-level test
suite parallelism. It's our hope that with these changes, we can get
the runtime of the integration test suite down to 20 minutes or so.
-UnifiedGenotyper tests: these divided nicely into logical categories
that also happened to distribute the runtime fairly evenly
-UnifiedGenotyperPloidy: these had to be divided arbitrarily into two
classes in order to halve the runtime
-HaplotypeCaller: turns out that the tests for complex and symbolic
variants make up half the runtime here, so merely moving these into
a separate class was sufficient
-BiasedDownsampling: most of these tests use excessively large intervals
that likely can't be reduced without defeating the goals of the tests. I'm
disabling these tests for now until they can either be redesigned to use smaller
intervals around the variants of interest, or refactored into unit tests
(creating a JIRA for Yossi for this task)
* Removed from codebase NestedHashMap since it is unused and untested.
* Integration tests change because the BQSR CSV is now sorted automatically.
* Resolves GSA-732
-replace unnecessary uses of the UnifiedGenotyper by public integration tests
with PrintReads
-move NanoSchedulerIntegrationTest to protected, since it's completely dependent
on the UnifiedGenotyper
-was previously set to 30, which seems far too aggressive given that with
ActiveRegionWalkers, as with LocusWalkers, this limits the depth of any
pileup returned by LIBS
-250 is a more conservative default used by the UG
-can adjust down/up later based on further experiments (GSA-699 will
remain open)
-verified with Ryan that all integration test differences are either
innocent or represent an improvement
GSA-699
The issue here is that the OptimizedLikelihoodTestProvider uses the same basic underlying class as the
BasicLikelihoodTestProvider and we were using the BasicTestProvider functionality to pull out tests of
that class; so if the optimized tests were run first we were unintentionally running those same tests
again with the basic ones (but expecting different results).
-- This is done to take advantage of longer reads which can produce less ambiguous haplotypes
-- Integration tests change for HC and BiasedDownsampling
-- Instead of doing a full SW alignment against the reference we read off bubbles from the assembly graph.
-- Smith-Waterman is run only on the base composition of the bubbles which drastically reduces runtime.
-- Refactoring graph functions into a new DeBruijnAssemblyGraph class.
-- Bug fix in path.getBases().
-- Adding validation code to the assembly engine.
-- Renaming SimpleDeBruijnAssembler to match the naming of the new Assembly graph class.
-- Adding bug fixes, docs and unit tests for DeBruijnAssemblyGraph and KBestPaths classes.
-- Added ability to ignore bubbles that are too divergent from the reference
-- Max kmer can't be bigger than the extension size.
-- Reverse the order that we create the assembly graphs so that the bigger kmers are used first.
-- New algorithm for determining unassembled insertions based on the bubble traversal instead of the full SW alignment.
-- Don't need the full read span reference loc for anything any more now that we clip down to the extended loc for both assembly and likelihood evaluation.
-- Updating HaplotypeCaller and BiasedDownsampling integration tests.
-- Rebased everything into one commit as requested by Eric
-- improvements to the bubble traversal are coming as a separate push
These 2 changes improve runtime performance almost as much as Ryan's previous attempt (with ID-based comparisons):
* Don't unnecessarily overload Allele.getBases() in the Haplotype class.
* Haplotype.getBases() was calling clone() on the byte array.
* Added a constructor to Allele (and Haplotype) that takes in an Allele as input.
* It makes a copy of he given allele without having to go through the validation of the bases (since the Allele has already been validated).
* Rev'ed the variant jar accordingly.
For the reviewer: all tests passed before rebasing, so this should be good to go as far as correctness.
-- modified ReadBin GenomeLoc to keep track of softStart() and softEnd() of the reads coming in, to make sure the reference will always be sufficient even if we want to use the soft-clipped bases
-- changed the verification from readLength to aligned bases to allow reads with soft-clipped bases
-- switched TreeSet -> PriorityQueue in the ConstrainedMateFixer as some different reads can be considered equal by picard's SAMRecordCoordinateComparator (the Set was replacing them)
-- pulled out ReadBin class so it can be testable
-- added unit tests for ReadBin with soft-clips
-- added tests for getMismatchCount (AlignmentUtils) to make sure it works with soft-clipped reads
GSA-774 #resolve
-- Active regions are created as normal, but they are split and trimmed to the engine intervals when added to the traversal, if there are intervals present.
-- UnitTests for ActiveRegion.splitAndTrimToIntervals
-- GenomeLocSortedSet.getOverlapping uses binary search to efficiently in ~ log N time find overlapping intervals
-- UnitTesting overlap function in GenomeLocSortedSet
-- Discovered fundamental implementation bug in that adding genome locs out of order (elements on 20 then on 19) produces an invalid GenomeLocSortedSet. Created a JIRA to address this: https://jira.broadinstitute.org/browse/GSA-775
-- Constructor that takes a collection of genome locs now sorts its input and merges overlapping intervals
-- Added docs for the constructors in GLSS
-- Update HaplotypeCaller MD5s, which change because ActiveRegions are now restricted to the engine intervals, which changes slightly the regions in the tests and so the reads in the regions, and thus the md5s
-- GenomeAnalysisEngineUnitTest needs to provide non-null genome loc parser
-- Increase the allowed runtime of one UG integration test
-- The GGA indels mode runs two UG commands, and was barely under the 10 minute limit before. Some updates can push this right over the edge. Increased limit
-- CalibrateGenotypeLikelihoods runs on a small data set now, so it's faster
-- Updating MD5s due to more correct quality utils. DuplicatesWalkers quality estimates have changed. One UG test has different FS and rank sum tests because the conversion to phred scores are slightly (second decimal place) different
-- The UG was using MathUtils binomial probability backward, so that the estimated confidence was always NaN, and was as a side effect other utils converted this to a meaningless 0.0. This is all because there wasn't a unit test.
-- I've fixed the calculation, so it's now log10 based, uses robust MathUtils and QualityUtils functions to compute probabilities, and added a unit test.
-- Fixed a few conversion bugs with edge case quals (ones that were very high)
-- Fixed a critical bug in the conversion of quals that was causing near capped quals to fall below their actual value. Will undoubtedly need to fix md5s
-- More precise prob -> qual calculations for very high confidence events in phredScaleCorrectRate, trueProbToQual, and errorProbToQual. Very likely to improve accuracy of many calculations in the GATK
-- Added errorProbToQual and trueProbToQual calculations that accept an integer cap, and perform the (tricky) conversion from int to byte correctly.
-- Full docs and unit tests for phredScaleCorrectRate and phredScaleErrorRate.
-- Renamed probToQual to trueProbToQual
-- Added goodProbability and log10OneMinusX to MathUtils
-- Went through the GATK and cleaned up many uses of QualityUtils
-- Cleanup constants in QualityUtils
-- Added full docs for all of the constants
-- Rename MAX_QUAL_SCORE to MAX_SAM_QUAL_SCORE for clarity
-- Moved MAX_GATK_USABLE_Q_SCORE to RecalDatum, as it's s BQSR specific feature
-- Convert uses of QualityUtils.errorProbToQual(1-x) to QualityUtils.trueProbToQual(x)
-- Cleanup duplicate quality score routines in MathUtils. Moved and renamed MathUtils.log10ProbabilityToPhredScale => QualityUtils.phredScaleLog10ErrorRate. Removed 3 routines from MathUtils, and remapped their usages into the better routines in QualityUtils
-- Renamed ValidatePileup to CheckPileup since validation is reserved word
-- Renamed AlignmentValidation to CheckAlignment (same as above)
-- Refactored category definitions to use constants defined in HelpConstants
-- Fixed a couple of minor typos and an example error
-- Reorganized the GATKDocs index template to use supercategories
-- Refactored integration tests for renamed walkers (my earlier refactoring had screwed them up or not carried over)
- got md5s from a interim version that does not have the per-sample downsampling hookedup
- added an integration test that forces the result from flat-downsampling to equal that which results from an equivalent flat contamination file
-- HaplotypeCaller and PerReadAlleleLikelihoodMap should use LinkedHashMaps instead of plain HashMaps. That way the ordering when traversing alleles is maintained. If the JVM traverses HashMaps with random ordering, different reads (with same likelihood) may be removed by contamination checker, and different alleles may be picked if they have same likelihoods for all reads.
-- Put in some GATKDocs and contracts in HaplotypeCaller files (far from done, code is a beast)
-- Update md5's due to different order of iteration in LinkedHashMaps instead of HashMaps inside HaplotypeCaller (due to change in PerReadAlleleLikelihoodMap that also slightly modifies reads chosen by per-read downsampling).
-- Reenabled testHaplotypeCallerMultiSampleGGAMultiAllelic test
-- Added some defensive argument checks into HaplotypeCaller public functions (not intended to be done yet).
-- Sorted out contents of BAM Processing vs. Diagnostics & QC Tools
-- Moved two validation-related walkers from Diagnostics & QC to Validation Utilities
-- Reworded some category names and descriptions to be more explicit and user-friendly
-- New HMM has two impacts on MD5s. First, all indel calls with UG and all calls by HC no longer have the HaplotypeScore computed. This is for the good, especially given the computational cost of this annotationa and unclear value for HC. Second, the BaseQualityRankSum values are changing by tiny amounts because of the changes in the HMM likelihoods.
-- Disabled three tests from Yossi that cause strange MD5 differences with calls for HC, created a JIRA for him to enable and fix
-- Disabled the non-deterministic GGA test. Assigned JIRA to Guillermo
-- With this push I expect all integration tests to pass
-- The new HMM new edge conditions the likelihoods are offset by log10(n possible starts) so the results don't really mean "fits the haplotype well" any longer. This results in grossly inflated HaplotypeScores for indels and with the HaplotypeCaller. So I'm simply not going to emit this annotation value any longer for indels and for the HC
-- Uses 1/N for N potential start sites as the probability of starting at any one of the potential start sites
-- Add flag that says to use the original edge condition, respected by all subclasses. This brings the new code back to the original state, but with all of the cleanup I've done
-- Only test configurations where the read length <= haplotype length. I think this is actually the contract, but we'll talk about this tomorrow
-- Fix egregious bug with the myLog10SumLog10 function doing the exact opposite of the requested arguments, so that doExact really meant don't do exact
-- PairHMM now exposes computeReadLikelihoodGivenHaplotypeLog10 but subclasses must overload subComputeReadLikelihoodGivenHaplotypeLog10. This protected function does the work, and the public function will do argument and result QC
-- Have to be more tolerant of reference (approximate) HMM. All unit tests from the original HMM implementations pass now
-- Added locs of docs
-- Generalize unit tests with multiple equivalent matches of read to haplotype
-- Added runtime argument checking for initial and computeReadLikelihoodGivenHaplotypeLog10
-- Functions to dumpMatrices for debugging
-- Fix nasty bug (without original unit tests) in LoglessPairHMM
-- Max read and haplotype lengths only worked in previous code if they were exactly equal to the provided read and haplotype sizes. Fixed bug. Added unit test to ensure this doesn't break again.
-- Added dupString(string, n) method to Utils
-- Added TODOs for next commit. Need to compute number of potential start sites not in initialize but in the calc routine since this number depends not on the max sizes but the actual read sizes
-- Unit tests for the hapStartIndex functionality of PairHMM
-- Moved computeFirstDifferingPosition to PairHMM, and added unit tests
-- Added extensive unit tests for the hapStartIndex functionality of computeReadLikelihoodGivenHaplotypeLog10
-- Still TODOs left in the code that I'll fix up
-- Logless now compute constants, if they haven't been yet initialized, even if you forgot to say so
-- General: the likelihood penalty for potential start sites is now properly computed against the actual read and reference bases, not the maximum. This involved moving some initialize() code into the computeLikelihoods function. That's ok because all of the potential log10 functions are actually going to cached versions, so the slowdown is minimal
-- Added some unit tests to ensure that common errors (providing haplotypes too long, reads too long, not initializing the HMM) are captured as errors
David Roazen
depristo
Eric Banks
Alec Wysoker
Ryan Poplin
Mauricio Carneiro
Guillermo del Angel
Tad Jordan
Geraldine Van der Auwera
Yossi Farjoun