We now run Smith-Waterman on the dangling tail against the corresponding reference tail.
If we can generate a reasonable, low entropy alignment then we trigger the merge to the
reference path; otherwise we abort. Also, we put in a check for low-complexity of graphs
and don't let those pass through.
Added tests for this implementation that checks exact SW results and correct edges added.
Principle is simple: when coverage is deep enough, any single-base read error will look like a rare k-mer but correct sequence will be supported by many reads to correct sequences will look like common k-mers. So, algorithm has 3 main steps:
1. K-mer graph buildup.
For each read in an active region, a map from k-mers to the number of times they have been seen is built.
2. Building correction map.
All "rare" k-mers that are sparse (by default, seen only once), get mapped to k-mers that are good (by default, seen at least 20 times but this is a CL argument), and that lie within a given Hamming distance (by default, =1). This map can be empty (i.e. k-mers can be uncorrectable).
3. Correction proposal
For each constituent k-mer of each read, if this k-mer is rare and maps to a good k-mer, get differing base positions in k-mer and add these to a list of corrections for each base in each read. Then, correct read at positions where correction proposal is unanimous and non-empty.
The algorithm defaults are chosen to be very stringent and conservative in the correction: we only try to correct singleton k-mers, we only look for good k-mers lying at Hamming distance = 1 from them, and we only correct a base in read if all correction proposals are congruent.
By default, algorithm is disabled but can be enabled in HaplotypeCaller via the -readErrorCorrect CL option. However, at this point it's about 3x-10x more expensive so it needs to be optimized if it's to be used.
Ns are treated as wildcards in the PairHMM so creating haplotypes with Ns gives them artificial advantages over other ones.
This was the cause of at least one FN where there were Ns at a SNP position.
Problem:
The sequence graphs can get very complex and it's not enough just to test that any given read has non-unique kmers.
Reads with variants can have kmers that match unique regions of the reference, and this causes cycles in the final
sequence graph. Ultimately the problem is that kmers of 10/25 may not be large enough for these complex regions.
Solution:
We continue to try kmers of 10/25 but detect whether cycles exist; if so, we do not use them. If (and only if) we
can't get usable graphs from the 10/25 kmers, then we start iterating over larger kmers until we either can generate
a graph without cycles or attempt too many iterations.
-- 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
- Memoized MathUtil's cumulative binomial probability function.
- Reduced the default size of the read name map in reduced reads and handle its resets more efficiently.
-- Created a new annotation DepthPerSampleHC that is by default on in the HaplotypeCaller
-- The depth for the HC is the sum of the informative alleles at this site. It's not perfect (as we cannot differentiate between reads that align over the event but aren't informative vs. those that aren't even close) but it's a pretty good proxy and it matches with the AD field (i.e., sum(AD) = DP).
-- Update MD5s
-- delivers [#48240601]
-- In the case where we have multiple potential alternative alleles *and* we weren't calling all of them (so that n potential values < n called) we could end up trimming the alleles down which would result in the mismatch between the PerReadAlleleLikelihoodMap alleles and the VariantContext trimmed alleles.
-- Fixed by doing two things (1) moving the trimming code after the annotation call and (2) updating AD annotation to check that the alleles in the VariantContext and the PerReadAlleleLikelihoodMap are concordant, which will stop us from degenerating in the future.
-- delivers [#50897077]
-- Ultimately this was caused by overly aggressive merging of CommonSuffixMerger. In the case where you have this graph:
ACT [ref source] -> C
G -> ACT -> C
we would merge into
G -> ACT -> C
which would linearlize into
GACTC
Causing us to add bases to the reference source node that couldn't be recovered. The solution was to ensure that CommonSuffixMerger only operates when all nodes to be merged aren't source nodes themselves.
-- Added a convenient argument to the haplotype caller (captureAssemblyFailureBAM) that will write out the exact reads to a BAM file that went into a failed assembly run (going to a file called AssemblyFailure.BAM). This can be used to rerun the haplotype caller to produce the exact error, which can be hard in regions of deep coverage where the downsampler state determines the exact reads going into assembly and therefore makes running with a sub-interval not reproduce the error
-- Did some misc. cleanup of code while debugging
-- [delivers #50917729]
-- The previous implementation attempted to be robust to this, but not all cases were handled properly. Added a helper function updateInde() that bounds up the update to be in the range of the indel array, and cleaned up logic of how the method works. The previous behavior was inconsistent across read fwd/rev stand, so that the indel cigars at the end of read were put at the start of reads if the reads were in the forward strand but not if they were in the reverse strand. Everything is now consistent, as can be seen in the symmetry of the unit tests:
tests.add(new Object[]{"1D3M", false, EventType.BASE_DELETION, new int[]{0,0,0}});
tests.add(new Object[]{"1M1D2M", false, EventType.BASE_DELETION, new int[]{1,0,0}});
tests.add(new Object[]{"2M1D1M", false, EventType.BASE_DELETION, new int[]{0,1,0}});
tests.add(new Object[]{"3M1D", false, EventType.BASE_DELETION, new int[]{0,0,1}});
tests.add(new Object[]{"1D3M", true, EventType.BASE_DELETION, new int[]{1,0,0}});
tests.add(new Object[]{"1M1D2M", true, EventType.BASE_DELETION, new int[]{0,1,0}});
tests.add(new Object[]{"2M1D1M", true, EventType.BASE_DELETION, new int[]{0,0,1}});
tests.add(new Object[]{"3M1D", true, EventType.BASE_DELETION, new int[]{0,0,0}});
tests.add(new Object[]{"4M1I", false, EventType.BASE_INSERTION, new int[]{0,0,0,1,0}});
tests.add(new Object[]{"3M1I1M", false, EventType.BASE_INSERTION, new int[]{0,0,1,0,0}});
tests.add(new Object[]{"2M1I2M", false, EventType.BASE_INSERTION, new int[]{0,1,0,0,0}});
tests.add(new Object[]{"1M1I3M", false, EventType.BASE_INSERTION, new int[]{1,0,0,0,0}});
tests.add(new Object[]{"1I4M", false, EventType.BASE_INSERTION, new int[]{0,0,0,0,0}});
tests.add(new Object[]{"4M1I", true, EventType.BASE_INSERTION, new int[]{0,0,0,0,0}});
tests.add(new Object[]{"3M1I1M", true, EventType.BASE_INSERTION, new int[]{0,0,0,0,1}});
tests.add(new Object[]{"2M1I2M", true, EventType.BASE_INSERTION, new int[]{0,0,0,1,0}});
tests.add(new Object[]{"1M1I3M", true, EventType.BASE_INSERTION, new int[]{0,0,1,0,0}});
tests.add(new Object[]{"1I4M", true, EventType.BASE_INSERTION, new int[]{0,1,0,0,0}});
-- delivers #50445353
-- We now inject the given alleles into the reference haplotype and add them to the graph.
-- Those paths are read off of the graph and then evaluated with the appropriate marginalization for GGA mode.
-- This unifies how Smith-Waterman is performed between discovery and GGA modes.
-- Misc minor cleanup in several places.
The problem ultimately was that ReadUtils.readStartsWithInsertion() ignores leading hard/softclips, but
ReduceReads does not. So I refactored that method to include a boolean argument as to whether or not
clips should be ignored. Also rebased so that return type is no longer a Pair.
Added unit test to cover this situation.
-- Started by Mark. Finished up by Ryan.
-- GGA mode still respected glm argument for SNP and INDEL models, so that you would silently fail to genotype indels at all if the -glm INDEL wasn't provided, but you'd still emit the sites, so you'd see records in the VCF but all alleles would be no calls.
-- https://www.pivotaltracker.com/story/show/48924339 for more information
-- [resolves#48924339]
Problem
--------
Diagnose Targets is outputting missing intervals to stdout if the argument -missing is not provided
Solution
--------
Make it NOT default to stdout
[Delivers #50386741]
BandedHMM
---------
-- An implementation of a linear runtime, linear memory usage banded logless PairHMM. Thought about 50% faster than current PairHMM, this implementation will be superceded by the GraphHMM when it becomes available. The implementation is being archived for future reference
Useful infrastructure changes
-----------------------------
-- Split PairHMM into a N2MemoryPairHMM that allows smarter implementation to not allocate the double[][] matrices if they don't want, which was previously occurring in the base class PairHMM
-- Added functionality (controlled by private static boolean) to write out likelihood call information to a file from inside of LikelihoodCalculationEngine for using in unit or performance testing. Added example of 100kb of data to private/testdata. Can be easily read in with the PairHMMTestData class.
-- PairHMM now tracks the number of possible cell evaluations, and the LoglessCachingPairHMM updates the nCellsEvaluated so we can see how many cells are saved by the caching calculation.
-- Previous version took a Collection<GATKSAMRecord> to remove, and called ArrayList.removeAll() on this collection to remove reads from the ActiveRegion. This can be very slow when there are lots of reads, as ArrayList.removeAll ultimately calls indexOf() that searches through the list calling equals() on each element. New version takes a set, and uses an iterator on the list to remove() from the iterator any read that is in the set. Given that we were already iterating over the list of reads to update the read span, this algorithm is actually simpler and faster than the previous one.
-- Update HaplotypeCaller filterReadsInRegion to use a Set not a List.
-- Expanded the unit tests a bit for ActiveRegion.removeAll
Bug fixes and missing interval functionality for Diagnose Targets
While the code seems fine, the complex parts of it are untested. This is probably fine for now, but private code can have a tendency to creep into the codebase once accepted. I would have preferred that unit test OR a big comment stating that the code is untested (and thus broken by Mark's rule).
It is with these cavets that I accept the pull request.
Problem
------
Diagnose Targets identifies holes in the coverage of a targetted experiment, but it only reports them doesn't list the actual missing loci
Solution
------
This commit implements an optional intervals file output listing the exact loci that did not pass filters
Itemized changes
--------------
* Cache callable statuses (to avoid recalculation)
* Add functionality to output missing intervals
* Implement new tool to qualify the missing intervals (QualifyMissingIntervals) by gc content, size, type of missing coverage and origin (coding sequence, intron, ...)
Problem
-------
When the interval had no reads, it was being sent to the VCF before the intervals that just got processed, therefore violating the sort order of the VCF.
Solution
--------
Use a linked hash map, and make the insertion and removal all happen in one place regardless of having reads or not. Since the input is ordered, the output has to be ordered as well.
Itemized changes
--------------
* Clean up code duplication in LocusStratification and SampleStratification
* Add number of uncovered sites and number of low covered sites to the VCF output.
* Add new VCF format fields
* Fix outputting multiple status when threshold is 0 (ratio must be GREATER THAN not equal to the threshold to get reported)
[fixes#48780333]
[fixes#48787311]
-- 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.
* bitset could legitimately be in an unfinished state but we were trying to access it without finalizing.
* added --cancer_mode argument per Mark's suggestion to force the user to explicitly enable multi-sample mode.
* tests were easiest to implement as integration tests (this was a really complicated case).
Problem
-------
The DeBruijn assembler was too slow. The cause of the slowness was the need to construct many kmer graphs (from max read length in the interval to 11 kmer, in increments of 6 bp). This need to build many kmer graphs was because the assembler (1) needed long kmers to assemble through regions where a shorter kmer was non-unique in the reference, as we couldn't split cycles in the reference (2) shorter kmers were needed to be sensitive to differences from the reference near the edge of reads, which would be lost often when there was chain of kmers of longer length that started before and after the variant.
Solution
--------
The read threading assembler uses a fixed kmer, in this implementation by default two graphs with 10 and 25 kmers. The algorithm operates as follows:
identify all non-unique kmers of size K among all reads and the reference
for each sequence (ref and read):
find a unique starting position of the sequence in the graph by matching to a unique kmer, or starting a new source node if non exist
for each base in the sequence from the starting vertex kmer:
look at the existing outgoing nodes of current vertex V. If the base in sequence matches the suffix of outgoing vertex N, read the sequence to N, and continue
If no matching next vertex exists, find a unique vertex with kmer K. If one exists, merge the sequence into this vertex, and continue
If a merge vertex cannot be found, create a new vertex (note this vertex may have a kmer identical to another in the graph, if it is not unique) and thread the sequence to this vertex, and continue
This algorithm has a key property: it can robustly use a very short kmer without introducing cycles, as we will create paths through the graph through regions that aren't unique w.r.t. the sequence at the given kmer size. This allows us to assemble well with even very short kmers.
This commit includes many critical changes to the haplotype caller to make it fast, sensitive, and accurate on deep and shallow WGS and exomes, the key changes are highlighted below:
-- The ReadThreading assembler keeps track of the maximum edge multiplicity per sample in the graph, so that we prune per sample, not across all samples. This change is essential to operate effectively when there are many deep samples (i.e., 100 exomes)
-- A new pruning algorithm that will only prune linear paths where the maximum edge weight among all edges in the path have < pruningFactor. This makes pruning more robust when you have a long chain of bases that have high multiplicity at the start but only barely make it back into the main path in the graph.
-- We now do a global SmithWaterman to compute the cigar of a Path, instead of the previous bubble-based SmithWaterman optimization. This change is essential for us to get good variants from our paths when the kmer size is small. It also ensures that we produce a cigar from a path that only depends only the sequence of bases in the path, unlike the previous approach which would depend on both the bases and the way the path was decomposed into vertices, which depended on the kmer size we used.
-- Removed MergeHeadlessIncomingSources, which was introducing problems in the graphs in some cases, and just isn't the safest operation. Since we build a kmer graph of size 10, this operation is no longer necessary as it required a perfect match of 10 bp to merge anyway.
-- The old DebruijnAssembler is still available with a command line option
-- The number of paths we take forward from the each assembly graph is now capped at a factor per sample, so that we allow 128 paths for a single sample up to 10 x nSamples as necessary. This is an essential change to make the system work well for large numbers of samples.
-- Add a global mismapping parameter to the HC likelihood calculation: The phredScaledGlobalReadMismappingRate reflects the average global mismapping rate of all reads, regardless of their mapping quality. This term effects the probability that a read originated from the reference haploytype, regardless of its edit distance from the reference, in that the read could have originated from the reference haplotype but from another location in the genome. Suppose a read has many mismatches from the reference, say like 5, but has a very high mapping quality of 60. Without this parameter, the read would contribute 5 * Q30 evidence in favor of its 5 mismatch haplotype compared to reference, potentially enough to make a call off that single read for all of these events. With this parameter set to Q30, though, the maximum evidence against the reference that this (and any) read could contribute against reference is Q30. -- Controllable via a command line argument, defaulting to Q60 rate. Results from 20:10-11 mb for branch are consistent with the previous behavior, but this does help in cases where you have rare very divergent haplotypes
-- Reduced ActiveRegionExtension from 200 bp to 100 bp, which is a performance win and the large extension is largely unnecessary with the short kmers used with the read threading assembler
Infrastructure changes / improvements
-------------------------------------
-- Refactored BaseGraph to take a subclass of BaseEdge, so that we can use a MultiSampleEdge in the ReadThreadingAssembler
-- Refactored DeBruijnAssembler, moving common functionality into LocalAssemblyEngine, which now more directly manages the subclasses, requiring them to only implement a assemble() method that takes ref and reads and provides a List<SeqGraph>, which the LocalAssemblyEngine takes forward to compute haplotypes and other downstream operations. This allows us to have only a limited amount of code that differentiates the Debruijn and ReadThreading assemblers
-- Refactored active region trimming code into ActiveRegionTrimmer class
-- Cleaned up the arguments in HaplotypeCaller, reorganizing them and making arguments @Hidden and @Advanced as appropriate. Renamed several arguments now that the read threading assembler is the default
-- LocalAssemblyEngineUnitTest reads in the reference sequence from b37, and assembles with synthetic reads intervals from 10-11 mbs with only the reference sequence as well as artificial snps, deletions, and insertions.
-- Misc. updates to Smith Waterman code. Added generic interface to called not surpisingly SmithWaterman, making it easier to have alternative implementations.
-- Many many more unit tests throughout the entire assembler, and in random utilities
* This is emerging now because BWA-MEM produces lots of reads that are not primary alignments
* The ConstrainedMateFixingManager class used by IndelRealigner was mis-adjusting SAM flags because it
was getting confused by these secondary alignments
* Added unit test to cover this case
RR counts are represented as offsets from the first count, but that wasn't being done
correctly when counts are adjusted on the fly. Also, we were triggering the expensive
conversion and writing to binary tags even when we weren't going to write the read
to disk.
The code has been updated so that unconverted counts are passed to the GATKSAMRecord
and it knows how to encode the tag correctly. Also, there are now methods to write
to the reduced counts array without forcing the conversion (and methods that do force
the conversion).
Also:
1. counts are now maintained as ints whenever possible. Only the GATKSAMRecord knows
about the internal encoding.
2. as discussed in meetings today, we updated the encoding so that it can now handle
a range of values that extends to 255 instead of 127 (and is backwards compatible).
3. tests have been moved from SyntheticReadUnitTest to GATKSAMRecordUnitTest accordingly.
-- Added check to see if read spans beyond reference window MINUS padding and event length. This guarantees that read will always be contained in haplotype.
-- Changed md5's that happen when long reads from old 454 data have their likelihoods changed because of the extra base clipping.
Use case:
The default AF priors used (infinite sites model, neutral variation) is appropriate in the case where the reference allele is ancestral, and the called allele is a derived allele.
Most of the times this is true but in several population studies and in ancient DNA analyses this might introduce reference biases, and in some other cases it's hard to ascertain what the ancestral allele is (normally requiring to look up homologous chimp sequence).
Specifying no prior is one solution, but this may introduce a lot of artifactual het calls in shallower coverage regions.
With this option, users can specify what the prior for each AC should be according to their needs, subject to the restrictions documented in the code and in GATK docs.
-- Updated ancient DNA single sample calling script with filtering options and other cleanups.
-- Added integration test. Removed old -noPrior syntax.
Note that this works only in the case of pileups (i.e. coming from UG);
allele-biased down-sampling for RR just cannot work for haplotypes.
Added lots of unit tests for new functionality.
1. Using cumulative binomial probability was not working at high coverage sites (because p-values quickly
got out of hand) so instead we use a hybrid system for determining significance: at low coverage sites
use binomial prob and at high coverage sites revert to using the old base proportions. Then we get the
best of both worlds. As a note, coverage refers to just the individual base counts and not the entire pileup.
2. Reads were getting lost because of the comparator being used in the SlidingWindow. When read pairs had
the same alignment end position the 2nd one encountered would get dropped (but added to the header!). We
now use a PriorityQueue instead of a TreeSet to allow for such cases.
3. Each consensus keeps track of its own number of softclipped bases. There was no reason that that number
should be shared between them.
4. We output consensus filtered (i.e. low MQ) reads whenever they are present for now. Don't lose that
information. Maybe we'll decide to change this in the future, but for now we are conservative.
5. Also implemented various small performance optimizations based on profiling.
Added unit tests to cover these changes; systematic assessment now tests against low MQ reads too.
Mark DePristo
Ryan Poplin
Eric Banks
Guillermo del Angel
Mauricio Carneiro
Michael McCowan
Yossi Farjoun