Initial test to see how Bamboo will respond. More detailed email to follow.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4609 348d0f76-0448-11de-a6fe-93d51630548a
Unrelated change: in the sorted-target mode (when we read sorted target intervals one by on from a file), one can now specify multiple semicolon-separated interval files (all must be sorted). Not hugely useful probably, but makes --targetIntervals always process its values in exactly the same way, so we are consistent (it has been already taking ;-separated args in unsorted mode)
NwayIntervalMergingIterator: reads in multiple sorted GenomeLoc input streams (iterators) and presents them as a single sorted and merged stream
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4602 348d0f76-0448-11de-a6fe-93d51630548a
a) Sites were numerically well behaved now, but another hard fact of life is that the AF iteration is defined in linear Pr space, not in log likelihood space, and the math doesn't work out in log space. So, we need to convert back and forth from lin to log space.
b) As a consequence of a), the code got a major slowdown, and calling the 629 samples was about 15 times slower than before (sic).
c) To solve b), log10 of integers are now cached at init, and numerical approximations are now made. Most importantly, I'm using the approximation that log(exp(a) + exp(b)) ~= max(a,b) which seems almost inconsequential in practical performance but reduces computation time to what it was before. More detailes analyses are forthcoming. This approximation can be refined further on to avoid expensive log-exp conversions if further profiling and analysis deems it necessary.
Also, two other issues were solved:
a) Strand bias computation was actually wrong in the case where the optimal AC was bigger than max(forward reads,reverse reads). Now the code is exactly as buggy as the grid search model (all bugs are equal, but some are more equal than others)
b) Genotype likelihoods are now computed in a better way and if a likelihood < 0 we don't just cap to 0 but do something a bit smarter.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4600 348d0f76-0448-11de-a6fe-93d51630548a
interaction with Tribble. In Tribble, keeping these references in memory until
the shard is flushed means keeping one 512K character buffer per object in
memory. Fixed by purging the reference to the object at the end of the
shard traversal.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4599 348d0f76-0448-11de-a6fe-93d51630548a
of traversal to avoid holding a reference to the microscheduler, which holds a reference to
the engine, which in turn holds a reference to the walker, which itself holds a reference to
all the data aggregated during the course of the traversal.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4594 348d0f76-0448-11de-a6fe-93d51630548a
Updated a call to swapExt to specify the directory.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4586 348d0f76-0448-11de-a6fe-93d51630548a
parsing engine. Hugely lowers our memory footprint in integrationtests, but not yet enough to
run Mark's new parallelized VariantEvalIntegrationTests.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4585 348d0f76-0448-11de-a6fe-93d51630548a
- Forcing user to set the temp directory via -Djava.io.tmpdir to avoid filling up /tmp.
- By default deleting job outputs tagged as intermediate.
- Defaulting pipeline to scatter count 1 (no reads deleted).
- Cleaning up temp classes even when scripting fails.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4573 348d0f76-0448-11de-a6fe-93d51630548a
with -nt option. When last I checked in, Ryan was seeing a ~25% speedup
per shard by not indexing.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4556 348d0f76-0448-11de-a6fe-93d51630548a
with a more sensible strategy for sharding w/o BAMs at some point after
ASHG.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4549 348d0f76-0448-11de-a6fe-93d51630548a
a) Fix it up because it broke with a recent checkin to annotate vcf with unfiltered depth.
b) Printout of ref/alt alleles in output vcf was incorrect because the start/stop positions of associated GenomeLoc were incorrectly computed in case of a deletion.
c) Redid Beagle input/output walkers as not assume that ref was a single base, not to assume that variant was a vcf and generalized it to be indel-capable, so now the Beagle walkers can be used for indels as well.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4541 348d0f76-0448-11de-a6fe-93d51630548a
Queue GATK generated .intervals is now a List(File) again removing special case handling in the generator.
Instead of using @Scatter annotation, using ScatterFunction instance to determine if a job can be scattered.
Implemented special VcfGatherFunction which only uses the header from the first file, even if the other files differ in their headers.
Added a -deleteIntermediates to Queue to delete the outputs from intermediate commands after a successful run.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4536 348d0f76-0448-11de-a6fe-93d51630548a
some subset need
A C 1/1 --> C A 0/0
while another subset need
A C 1/1 --> C A 1/1
it's unclear how big these subsets are (or even if one is empty). What I do know is, doing the first one totally screws up concordance metrics for the 421-sample chip. So either something else needs to be done, or there's a bug in this walker. Until I know for sure, I've added an initialize exception to disable this thing...
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4523 348d0f76-0448-11de-a6fe-93d51630548a
forgot to add the samples to the header. How could the VCFWriter let me get away with something so boneheaded?!
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4513 348d0f76-0448-11de-a6fe-93d51630548a
Modified - SelectVariants: Hook up to VariantContextUtils to recalculate AC/AF/AN, which uses the accessor in VariantContext to do this. Somehow sites that were selected down to hom-ref genotypes only wound up getting positive AC.
**IMPORTANT** I kind of need input here. The header of a file used for an integration test specifies AC as being an integer. Recalculating it casts it into an integer list (which it should be, as it allows for alternate alleles). However this appears to clash with what the jexl expression is looking for? For now, the integration test itself needed to be changed -- it's unclear what to do when the header specifies AC of being one class, but recalculating it casts to another class, and I'm not sure what to do.
I'm committing my omni_qc pipeline because I'm almost certain 2 months down the road I'm going to wonder what the heck I did to generate my results.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4511 348d0f76-0448-11de-a6fe-93d51630548a
1: -sample can now include a file, which will be parsed for sample-name entries
2: If you request a sample to run analysis on, but it is not present in any of your RODs, VEW will exception out
3: Change added to parse Integer, String, and List<Integer> type Allele Count annotations (error otherwise)
4 [slightly problematic]: The count objects now maintain row-keys in order, as the keys were taking an inordinate amount of time in onTraversalDone (multiple calls to getRowKeys(), so many multiple sorts of the same underlying unsorted object, very bad)
There is a legacy comparison object which is unused which I will strip out soon.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4502 348d0f76-0448-11de-a6fe-93d51630548a
The core walker has been modified so that when variant contexts (eval and comp) are subset to command-line-specified sample(s), the chromosome count annotations (AC/AN/AF) are altered to reflect the AC/AN/AF of only those samples involved in the comparison. No more getting AC500 when you're comparing a 10-sample overlap. Interestingly enough, this didn't break any integration tests.
GenotypeConcordance now has two additional tables: Allele Count Statistics, and Allele Count Summary Statistics. These work exactly identically to the Sample Statistics and Sample Summary Statistics tables, except that the partition being used is no longer the sample, but instead the allele count of the variant sites. These tables stratify by both eval and comp ACs, e.g.
evalAC0
evalAC1
evalAC2
compAC0
compAC1
compAC2
Differences with previous integration tests were verified to only be in the Allele Count tables (by grepping them out of the diff); a new test has been added for the simple case of an AC=1 site in the eval becoming an AC=2 site in the comp.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4491 348d0f76-0448-11de-a6fe-93d51630548a
by the fact that the GATKSAMRecord, by design, needs to both inherit from
SAMRecord and wrap a 'member' SAMRecord, and method calls that aren't
implemented as explicit passthroughs can compromise the content of the
SAMRecord in subtle ways.
Will be automatically fixed when Picard moves to a lightweight SAMRecord
interface rather than the current heavyweight implementation. But in
the short-term, there's no obvious fix.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@4489 348d0f76-0448-11de-a6fe-93d51630548a
hanna
asivache
fromer
depristo
delangel
ebanks
kshakir
rpoplin
chartl