Complete re-write of PoolCaller algorithm, now basically beta quality code.
Improvements over PoolCaller include:
- more correct strand test
- fractional counts from genotypes (which means no individual lod threshold needed)
- signifigantly cleaner code; first beta-quality code I've written since BaitDesigner so long ago.
- faster, less likely to crash!
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@1020 348d0f76-0448-11de-a6fe-93d51630548a
1. Added logGamma function to utils
2. Required asserts to be enabled in the allele caller (run with java -ea)
3. put checks and asserts of NaN and Infinity in AlleleFrequencyEstimate
4. Added option FRACTIONAL_COUNTS to the pooled caller (not working right yet)
AlleleFrequencyWalker:
5. Made FORCE_1BASE_PROBS not static in AlleleFrequencyWalker (an argument should never be static! Jeez.)
6. changed quality_precision to be 1e-4 (Q40)
7. don't adjust by quality_precision unless the qual is actually zero.
8. added more asserts for NaN and Infinity
9. put in a correction for zero probs in P_D_q
10. changed pG to be hardy-weinberg in the presence of an allele frequency prior (duh)
11. rewrote binomialProb() to not overflow on deep coverage
12. rewrote nchoosek() to behave right on deep coverage
13. put in some binomailProb() tests in the main() routine (they come out right when compared with R)
Hunt for loci where 4bp should change things:
14. added FindNonrandomSecondBestBasePiles walker.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@471 348d0f76-0448-11de-a6fe-93d51630548a
Quals coming soon (four-base)
QHAT : Most likely alt allele freq (unconstrained by number of chromosomes).
QSTAR : Most likely alt allele freq (constrained by number of chromosomes).
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@402 348d0f76-0448-11de-a6fe-93d51630548a
EVEN MORE verbosity to come!
Tremble in anticipation.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@382 348d0f76-0448-11de-a6fe-93d51630548a
AlleleFrequencyWalker and related classes work equally well for 2 or 200 chromosomes.
Single Sample Calling:
Allele Frequency Metrics (LOD >= 5)
-------------------------------------------------
Total loci : 171575
Total called with confidence : 168615 (98.27%)
Number of variants : 111 (0.07%) (1/1519)
Fraction of variant sites in dbSNP : 87.39%
-------------------------------------------------
Hapmap metrics are coming up all zero. Will fix.
Pooled Calling:
AAF r-squared after EM is 0.99.
AAF r-squared after EM for alleles < 20% (in pools of ~100-200 chromosomes) is 0.95 (0.75 before EM)
Still not using fractional genotype counts in EM. That should improve r-squared for low frequency alleles.
Chores still outstanding:
- make a real pooled caller walker (as opposed to my experiment framework).
- add fractional genotype counts to EM cycle.
- add pool metrics to the metrics class? *shrug* we don't really have truth outside of a contrived experiment...
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@380 348d0f76-0448-11de-a6fe-93d51630548a
- AlleleMetrics and AlleleMetricrsWalker are now ready to take a list of clasess that implement the AllelicVariant interface
- Switched a genome location in AlleleFrequencyEstimate from String to GenomeLoc which makes way more sense.
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@280 348d0f76-0448-11de-a6fe-93d51630548a
- output GFF lines to a file (specified by a command line argument)
- improve the GFF output string
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@240 348d0f76-0448-11de-a6fe-93d51630548a
still need to add printing GFF intervals for stretches of confident reference calls.
does the GFF ROD class handle intervals?? We'll find out. >:)
git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@225 348d0f76-0448-11de-a6fe-93d51630548a
aaron
andrewk
kiran
jmaguire
depristo
hanna