Problem:
-Downsamplers were treating reduced reads the same as normal reads,
with occasionally catastrophic results on variant calling when an
entire reduced read happened to get eliminated.
Solution:
-Since reduced reads lack the information we need to do position-based
downsampling on them, best available option for now is to simply
exempt all reduced reads from elimination during downsampling.
Details:
-Add generic capability of exempting items from elimination to
the Downsampler interface via new doNotDiscardItem() method.
Default inherited version of this method exempts all reduced reads
(or objects encapsulating reduced reads) from elimination.
-Switch from interfaces to abstract classes to facilitate this change,
and do some minor refactoring of the Downsampler interface (push
implementation of some methods into the abstract classes, improve
names of the confusing clear() and reset() methods).
-Rewrite TAROrderedReadCache. This class was incorrectly relying
on the ReservoirDownsampler to preserve the relative ordering of
items in some circumstances, which was behavior not guaranteed by
the API and only happened to work due to implementation details
which no longer apply. Restructured this class around the assumption
that the ReservoirDownsampler will not preserve relative ordering
at all.
-Add disclaimer to description of -dcov argument explaining that
coverage targets are approximate goals that will not always be
precisely met.
-Unit tests for all individual downsamplers to verify that reduced
reads are exempted from elimination
-- Reuse infrastructure for RODs for reads to implement general IntervalReferenceOrderedView so that both TraverseReads and TraverseActiveRegions can use the same underlying infrastructure
-- TraverseActiveRegions now provides a meaningful RefMetaDataTracker to ActiveRegionWalker.map
-- Cleanup misc. code as it came up
-- Resolves GSA-808: Write general utility code to do rsID allele matching, hook up to UG and HC
-- Previously we used the LocusShardBalancer for the haplotype caller, which meant that TraverseActiveRegions saw its shards grouped in chunks of 16kb bits on the genome. These locus shards are useful when you want to use the HierarchicalMicroScheduler, as they provide fine-grained accessed to the underlying BAM, but they have two major drawbacks (1) we have to fairly frequently reset our state in TAR to handle moving between shard boundaries and (2) with the nano scheduled TAR we end up blocking at the end of each shard while our threads all finish processing.
-- This commit changes the system over to using an ActiveRegionShardBalancers, that combines all of the shard data for a single contig into a single combined shard. This ensures that TAR, and by extensions the HaplotypeCaller, gets all of the data on a single contig together so the the NanoSchedule runs efficiently instead of blocking over and over at shard boundaries. This simple change allows us to scale efficiently to around 8 threads in the nano scheduler:
-- See https://www.dropbox.com/s/k7f280pd2zt0lyh/hc_nano_linear_scale.pdf
-- See https://www.dropbox.com/s/fflpnan802m2906/hc_nano_log_scale.pdf
-- Misc. changes throughout the codebase so we Use the ActiveRegionShardBalancer where appropriate.
-- Added unit tests for ActiveRegionShardBalancer to confirm it does the merging as expected.
-- Fix bad toString in FilePointer
-- Made CountReadsInActiveRegions Nano schedulable, confirming identical results for linear and nano results
-- Made Haplotype NanoScheduled, requiring misc. changes in the map/reduce type so that the map() function returns a List<VariantContext> and reduce actually prints out the results to disk
-- Tests for NanoScheduling
-- CountReadsInActiveRegionsIntegrationTest now does NCT 1, 2, 4 with CountReadsInActiveRegions
-- HaplotypeCallerParallelIntegrationTest does NCT 1,2,4 calling on 100kb of PCR free data
-- Some misc. code cleanup of HaplotypeCaller
-- Analysis scripts to assess performance of nano scheduled HC
-- In order to make the haplotype caller thread safe we needed to use an AtomicInteger for the class-specific static ID counter in SeqVertex and MultiDebrujinVertex, avoiding a race condition where multiple new Vertex() could end up with the same id.
-- Add a maximum per sample and overall maximum number of reads held in memory by the ART at any one time. Does this in a new TAROrderedReadCache data structure that uses a reservior downsampler to limit the total number of reads to a constant amount. This constant is set to be by default 3000 reads * nSamples to a global maximum of 1M reads, all controlled via the ActiveRegionTraversalParameters annotation.
-- Added an integration test and associated excessively covered BAM excessiveCoverage.1.121484835.bam (private/testdata) that checks that the system is operating correctly.
-- #resolves GSA-921
-Make MaxRuntimeIntegrationTest more lenient by assuming that startup overhead
might be as long as 120 seconds on a very slow node, rather than the original
assumption of 20 seconds
-In TraverseActiveRegionsUnitTest, write temp bam file to the temp directory, not
to the current working directory
-SimpleTimerUnitTest: This test was internally inconsistent. It asserted that
a particular operation should take no more than 10 milliseconds, and then asserted
again that this same operation should take no more than 100 microseconds (= 0.1 millisecond).
On a slow node it could take slightly longer than 100 microseconds, however.
Changed the test to assert that the operation should require no more than 10000 microseconds
(= 10 milliseconds)
-change global default test timeout from 20 to 40 minutes (things just take longer
on the farm!)
-build.xml: allow runtestonly target to work with scala test classes
With LegacyLocusIteratorByState deleted, the legacy downsampling implementation
was already non-functional. This commit removes all remaining code in the
engine belonging to the legacy implementation.
-- The incremental version now processes active regions as soon as they are ready to be processed, instead of waiting until the end of the shard as in the previous version. This means that ART walkers will now take much less memory than previously. On chr20 of NA12878 the majority of regions are processed with as few as 500 reads in memory. Over the whole chr20 only 5K reads were ever held in ART at one time.
-- Fixed bug in the way active regions worked with shard boundaries. The new implementation no longer see shard boundaries in any meaningful way, and that uncovered a problem that active regions were always being closed across shard boundaries. This behavior was actually encoded in the unit tests, so those needed to be updated as well.
-- Changed the way that preset regions work in ART. The new contract ensures that you get exactly the regions you requested. the isActive function is still called, but its result has no impact on the regions. With this functionality is should be possible to use the HC as a generic assembly by forcing it to operate over very large regions
-- Added a few misc. useful functions to IncrementalActivityProfile
-- Required before I jump in an redo the entire activity profile so it's can be run imcrementally
-- This restructuring makes the differences between the two functionalities clearer, as almost all of the functionality is in the base class. The only functionality provided by the BandPassActivityProfile is isolated to a finalizeProfile function overloaded from the base class.
-- Renamed ActivityProfileResult to ActivityProfileState, as this is a clearer indication of its actual functionality. Almost all of the misc. walker changes are due to this name update
-- Code cleanup and docs for TraverseActiveRegions
-- Expanded unit tests for ActivityProfile and ActivityProfileState
-- UnitTests now include combinational tiling of reads within and spanning shard boundaries
-- ART now properly handles shard transitions, and does so efficiently without requiring hash sets or other collections of reads
-- Updating HC and CountReadsInActiveRegions integration tests
-Switch back to the old implementation, if needed, with --use_legacy_downsampler
-LocusIteratorByStateExperimental becomes the new LocusIteratorByState, and
the original LocusIteratorByState becomes LegacyLocusIteratorByState
-Similarly, the ExperimentalReadShardBalancer becomes the new ReadShardBalancer,
with the old one renamed to LegacyReadShardBalancer
-Performance improvements: locus traversals used to be 20% slower in the new
downsampling implementation, now they are roughly the same speed.
-Tests show a very high level of concordance with UG calls from the previous
implementation, with some new calls and edge cases that still require more examination.
-With the new implementation, can now use -dcov with ReadWalkers to set a limit
on the max # of reads per alignment start position per sample. Appropriate value
for ReadWalker dcov may be in the single digits for some tools, but this too
requires more investigation.
David Roazen
Mark DePristo
Eric Banks
Mauricio Carneiro
Joel Thibault