SGA updated to include new proportion-based insert size test.
Major fix for dichotomization test: MathUtils now optionally ignores NaN values for sums, averages, variances. In the future this feature can be pushed back into the AssociationContext object iself (e.g. no data? no entry), but it's kept like this for transparency for now.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5618 348d0f76-0448-11de-a6fe-93d51630548a
Model is disabled by default and we're still using the old Dindel model until I'm more confident that new model is a definitive improvement, so right now this is enabled by hidden command line arguments, and it's not to be used yet.
In detail:
a) Several refactorings to share softMax() available to other modules, so its now part of MathUtils.
b) Refactored a couple of read utilities and moved from BAQ to ReadUtils.
c) New PairHMMIndelErrorModel class implementing new likelihood model
d) Several new hidden debug arguments in UAC.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5389 348d0f76-0448-11de-a6fe-93d51630548a
- RefMetaDataTracker: grabbing variant contexts given a prefix (not sure where else this was implemented, if someone can show me I'll remove it)
- VCFUtils: grabbing VCF headers given a prefix
- MathUtils: Useful functions for calculating statistics on collections of Numbers
- VariantAnnotator: Made isUniqueHeaderLine a public static method -- maybe this should go into a different class. Not sure.
- Associations: PluginManager now used to propagate classes, implementations for Z,T,U tests, slight alteration to format to make the objects stored
in the window optionally different from those returned by whatever statistic is run across the window
Added:
- MannWhitneyU. Started to fix up WilcoxonRankSum but there are comments in there questioning the validity of some of the code, and I'm sure that
it's actually doing a U test. This implementation includes the direct calculation of p-values for small sample sizes, and a uniform approximation
for when one of the sample sets is small, and the other large. Unit tests to follow.
- BootstrapCallsMerger: takes n VCFs which have been called on the same samples; merges them together while averaging the annotations
- BootstrapCalls.q: qscript for testing the effectiveness of boostrap low-pass calling on the exome
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@5372 348d0f76-0448-11de-a6fe-93d51630548a
unicode quote characters embedded in it. These characters were invisible inside
IntelliJ but cause compile warnings for Ryan and Aaron, who for whatever reason
have a different default charset. Fixed.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@3203 348d0f76-0448-11de-a6fe-93d51630548a
Modified - Hapmap now takes a -q command to filter out variants by quality
Modified - MathUtils - cumBinomialProbLog now uses BigDecimal to handle some numerical imprecisions
Modified - PowerBelowFrequency - returns 0.0 if called with a negative number (can't be done from inside the walker itself, but since it's called elsewhere one can't be too careful)
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@2350 348d0f76-0448-11de-a6fe-93d51630548a
VariantAnnotator can be called as a standalone walker or by another walker, as it is by the UnifiedGenotyper. UG now no longer computes any of this meta data - it relegates the task completely to the annotator (assuming the output format accepts it).
This is a fairly all-encompassing check in. It involves changes to all of the UG code, bug fixes to much of the VCF code as things popped up, and other changes throughout. All integration tests pass and I've tediously confirmed that the annotation values are correct, but this framework could use some more rigorous testing.
Stage 2 of the process will happen later this week.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@2053 348d0f76-0448-11de-a6fe-93d51630548a
@MathUtils - added a new method: cumBinomialProbLog which calculates a cumulant from any start point to any end point using the BinomProbabilityLog calculation.
@PoolUtils - added a new utility class specifically for items related to pooled sequencing. A major part of the power calculation is now to calculate powers
independently by read direction. The only method in this class (currently) takes your reads and offsets, and splits them into two groups
by read direction.
@CoverageAndPowerWalker - completely rewritten to split coverage, median qualities, and power by read direction. Makes use of cumBinomialProbLog rather than
doing that calculation within the object itself.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@1462 348d0f76-0448-11de-a6fe-93d51630548a
binomialProbabilityLog uses a log-space calculation of the
binomial pmf to avoid the coefficient blowing up and thus
returning Infinity or NaN (or in some very strange cases
-Infinity). The log calculation compares very well, it seems
with our current method. It's in MathUtils but could stand
testing against rigorous truth data before becoming standard.
Added median calculator functions to ListUtils
getQScoreMedian is a new utility I wrote that given reads and
offsets will find the median Q score. While I was at it, I wrote
a similar method, getMedian, which will return the median of any
list of Comparables, independent of initial order. These are in
ListUtils.
Added a new poolseq directory and three walkers
CoverageAndPowerWalker is built on top of the PrintCoverage walker
and prints out the power to detect a mutant allele in a pool of
2*(number of individuals in the pool) alleles. It can be flagged
either to do this by boostrapping, or by pure math with a
probability of error based on the median Q-score. This walker
compiles, runs, and gives quite reasonable outputs that compare
visually well to the power calculation computed by Syzygy.
ArtificialPoolWalker is designed to take multiple single-sample
.bam files and create a (random) artificial pool. The coverage of
that pool is a user-defined proportion of the total coverage over
all of the input files. The output is not only a new .bam file,
but also an auxiliary file that has for each locus, the genotype
of the individuals, the confidence of that call, and that person's
representation in the artificial pool .bam at that locus. This
walker compiles and, uhh, looks pretty. Needs some testing.
AnalyzePowerWalker extends CoverageAndPowerWalker so that it can read previous power
calcuations (e.g. from Syzygy) and print them to the output file as well for direct
downstream comparisons.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@1460 348d0f76-0448-11de-a6fe-93d51630548a
Produces sane looking output on region of 1kG pilot1:
CALL NA12813.SRP000031.2009_02.bam CC 0.609084 0.609084
CALL NA12003.SRP000031.2009_02.bam CC 2.114234 2.114234 CCCCC
CALL NA06994.SRP000031.2009_02.bam CC 0.910114 0.910114 C
CALL NA18940.SRP000031.2009_02.bam CT 2.589749 0.910114 T
CALL NA18555.SRP000031.2009_02.bam CC 0.609084 0.609084
Next up, eval vs. Baseline pilot1 calls and pilot3 deep-coverage truth.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@525 348d0f76-0448-11de-a6fe-93d51630548a
rpoplin
delangel
chartl
ebanks
depristo
asivache
hanna
kiran
jmaguire