- Verified that exact calculations do agree with R's dwilcox()
- Verified that exact calculations do not agree with R's wilcox.test
+ This is because R does a correction, and calculates CDFs rather than PDFs (e.g. sums over dwilcox() values)
- Can now specify MWU to calculate cumulative exact tests, rather than point probabilities
- Z-scores are now calculated properly for exact tests
+ Previously, z-values calculated by inverting normal CDF from U-statistic PDF
+ Now both inversions are done, with a smart heuristic (biased variance) to make the point-calculated Z-value more accurate
+ Additional tests
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5911 348d0f76-0448-11de-a6fe-93d51630548a
Fixed minor problem with WalkerTest for "" (for parameterization) md5s.
Added an explicit integrationtest for BAQ NONE
Now only creates the BAQ'd pileup, if the useBAQPileup parameter is provide in initializeAlternateAllele.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5891 348d0f76-0448-11de-a6fe-93d51630548a
Also allows for the slightly more awesome name "MWUnitTest"
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5850 348d0f76-0448-11de-a6fe-93d51630548a
- Running a test when there are no observations of at least one of the sets now breaks the MWU contract
+ MWU returns Pair(Double.NaN,Double.NaN) in these instances to maintain the contract of never returning null
+ No more Double.Infinity values will appear
- RankSumTests now probe the return values for NaNs, and don't annotate if they appear
- For small sets where the probability is calculated recursively, the z-value is now the inversion of the error function
and not the approximate z-value
- UG and Annotator integration tests updated to reflect changes
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5845 348d0f76-0448-11de-a6fe-93d51630548a
Right now, if you select a multi-sample VCF file down (or one with filters I see) down to a smaller set of samples, and the site isn't polymorphic in that subgroup, then the alt allele is lost. For example, when selecting down NA12878 from the OMNI, I previously received the following VCF:
1 82154 rs4477212 A . . PASS AC=0;AF=0.00;AN=2;CR=100.0;DP=0;GentrainScore=0.7826;HW=1.0 GT:GC 0/0:0.7205
1 534247 SNP1-524110 C . . PASS AC=0;AF=0.00;AN=2;CR=99.93414;DP=0;GentrainScore=0.7423;HW=1.0 GT:GC 0/0:0.6491
1 565286 SNP1-555149 C T . PASS AC=2;AF=1.00;AN=2;CR=98.8266;DP=0;GentrainScore=0.7029;HW=1.0 GT:GC 1/1:0.3471
1 569624 SNP1-559487 T C . PASS AC=2;AF=1.00;AN=2;CR=97.8022;DP=0;GentrainScore=0.8070;HW=1.0 GT:GC 1/1:0.3942
Where the first two records lost the ALT allele, because NA12878 is hom-ref at this site. My change results in a VCF that looks like:
1 82154 rs4477212 A G . PASS AC=0;AF=0.00;AN=2;CR=100.0;DP=0;GentrainScore=0.7826;HW=1.0 GT:GC 0/0:0.7205
1 534247 SNP1-524110 C T . PASS AC=0;AF=0.00;AN=2;CR=99.93414;DP=0;GentrainScore=0.7423;HW=1.0 GT:GC 0/0:0.6491
1 565286 SNP1-555149 C T . PASS AC=2;AF=1.00;AN=2;CR=98.8266;DP=0;GentrainScore=0.7029;HW=1.0 GT:GC 1/1:0.3471
1 569624 SNP1-559487 T C . PASS AC=2;AF=1.00;AN=2;CR=97.8022;DP=0;GentrainScore=0.8070;HW=1.0 GT:GC 1/1:0.3942
The genotype remains unchanged, but the ALT allele is now preserved. I think this is the correct behavior, as reducing samples down shouldn't change the character of the site, only the AC in the subpopulation. This is related to the tricky issue of isPolymorphic() vs. isVariant().
isVariant => is there an ALT allele?
isPolymorphic => is some sample non-ref in the samples?
In part this is complicated as the semantics of sites-only VCFs, where ALT = . is used to mean not-polymorphic. Unfortunately, I just don't think there's a consistent convention right now, but it might be worth at some point to adopt a single approach to handling this. Wiki docs updated.
Does anyone have critical infrastructure that depends on the previous convention? Let me know so we can coordinate the change.
There's a new function subContextFromGenotypes() that also takes a Set<Allele> to handle this type of behavior.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5832 348d0f76-0448-11de-a6fe-93d51630548a
GenomeLocs can officially have any start/stop values from -Inf - +Inf. Bounds w.r.t. the reference are enforced, optionally, by GenomeLocParser. General code cleanup throughout the subsystem.
All validation code for GLs is now centralized, and all I/O systems now validate their inputs. Because of this, the Picard interval processing code has been changed to examine whether an interval is valid, and only keep the valid intervals. Note that the scatter/gather test was changed, because the original hg18 chr20 interval files as actually malformed (all records for some reason where on chr20).
Many interval processing routines were moved to IntervalUtils, as this is their natural home.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5830 348d0f76-0448-11de-a6fe-93d51630548a
Refactored several Interval utilties from GenomeLocParser to IntervalUtils, as one might expect they go
Removed GenomeLoc.clone() method, as this was not correctly implemented, and actually unnecessary, as GenomeLocs are immutable. Several iterator classes have changed to remove their use of clone()
Removed misc. unnecessary imports
Disabled, temporarily, the validating pileup integration test, as it uses reads mapped to an different reference sequence for ecoli, and this now does not satisfy the contracts for GenomeLoc
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5827 348d0f76-0448-11de-a6fe-93d51630548a
FindLargeShards. Runtime of FindLargeShards on papuan dataset is now 75min.
GATK proper should benefit as well, although the benefits might be so small
as to not be measurable.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5798 348d0f76-0448-11de-a6fe-93d51630548a
BAM files looking for outliers (outliers right now are defined naively as
shards whose sizes are more than 5 stddevs away from the mean). Runs in
13 minutes per chromosome on 707 low pass whole genome BAMs -- not great, but
much faster than running UG on the same region to discover anomalies.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5782 348d0f76-0448-11de-a6fe-93d51630548a
only download that test configuration when running unit/integration tests.
This means that the build will (hopefully) never break because it can't
fetch a file that isn't required for the GATK to run.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5775 348d0f76-0448-11de-a6fe-93d51630548a
a) Forgot to convert from phred to log-prob when computing gap penalties from recal table.
b) Forgot to uncomment code to correctly deal with hard-clipped bases in a read. But because of this, had to do a short term workaround to at least temporarily return class from hardClipAdaptorSequence to GATKSAMRecord. Otherwise, I get exceptions when casting because somehow some reads in HiSeq get to be SAMRecord (which GATKSAMRecord inherits from) but some reads get to be BAMRecords (which can't be cast into GATKSAMRecord), not sure why.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5771 348d0f76-0448-11de-a6fe-93d51630548a
Minor updates to the FCPTest to match the changes due to using the old indel caller.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5766 348d0f76-0448-11de-a6fe-93d51630548a
to extend from org.broadinstitute.sting.gatk.filters.ReadFilter rather than
directly from net.sf.picard.filter.SamRecordFilter, which allows us to add
an initialize(GATKEngine) method so that filters can do any initialization
they'd like based on CL arguments, SAM headers, etc.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5760 348d0f76-0448-11de-a6fe-93d51630548a
to the same standard as GATK format intervals. Full validation against contig
bounds is now performed for all intervals, regardless of their source. Also
fixed a few tests for validation exclusions that were backwards.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5698 348d0f76-0448-11de-a6fe-93d51630548a
Scattering non-contig interval lists by number of loci in the intervals instead of just number of intervals.
Queue caches the list of locs and how to split them up instead of reloading them from disk repeatedly.
TODO: general purpose function to divide data evenly.
Skip over comments when parsing picard analysis files.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5687 348d0f76-0448-11de-a6fe-93d51630548a
chartl
rpoplin
ebanks
depristo
delangel
hanna
kshakir
carneiro
droazen