Eric Banks d20a25d681 A much better way of choosing the alternate allele(s) to genotype in the SNP model of UG: instead of looking at the sum of base qualities (which can and did lead to us over-genotyping esp. when allowing multiple alternate alleles), we look at the likelihoods themselves (free since we are already calculating likelihoods for all 10 genotypes). Now, even if the base quals exceed some arbitrary threshold, we only bother genotyping an alternate allele when there's a sample for which it is more likely than ref/ref (I can generate weird edge cases where this falls apart, but none that model truly variable sites that we actually want to call). This leads to a huge efficiency improvement esp. for exomes (and esp. for many samples) where we almost always were trying to genotype all 3 alternate alleles. Integration tests change only because ref calls have slight QUAL differences (because the best alt allele is still chosen arbitrarily, but differently).		2011-12-27 16:50:38 -05:00
..
datasources	Better file count / buffering variation in test suite. Parameterized read shard buffering. Misc cleanup.	2011-12-06 10:10:38 -05:00
executive	Reorganized the codebase beneath top-level public and private directories,	2011-06-28 06:55:19 -04:00
filters	Reorganized the codebase beneath top-level public and private directories,	2011-06-28 06:55:19 -04:00
iterators	Better file count / buffering variation in test suite. Parameterized read shard buffering. Misc cleanup.	2011-12-06 10:10:38 -05:00
refdata	Merge with master	2011-11-19 09:56:06 -05:00
report	Right aligning GATKReport numeric columns and updated MD5s in tests.	2011-12-05 23:22:15 -05:00
samples	Added UnitTests for getFamilies() and getChildrenWithParents()	2011-12-09 15:57:07 +01:00
traversals	First pass of the asynchronous block loader.	2011-11-18 15:02:59 -05:00
walkers	A much better way of choosing the alternate allele(s) to genotype in the SNP model of UG: instead of looking at the sum of base qualities (which can and did lead to us over-genotyping esp. when allowing multiple alternate alleles), we look at the likelihoods themselves (free since we are already calculating likelihoods for all 10 genotypes). Now, even if the base quals exceed some arbitrary threshold, we only bother genotyping an alternate allele when there's a sample for which it is more likely than ref/ref (I can generate weird edge cases where this falls apart, but none that model truly variable sites that we actually want to call). This leads to a huge efficiency improvement esp. for exomes (and esp. for many samples) where we almost always were trying to genotype all 3 alternate alleles. Integration tests change only because ref calls have slight QUAL differences (because the best alt allele is still chosen arbitrarily, but differently).	2011-12-27 16:50:38 -05:00
EngineFeaturesIntegrationTest.java	Thanks to Mark for agreeing to transition from 'variants' back to 'variant'. I think I got them all but I've been jumping all around the code, so there might be a straggler or two.	2011-08-09 12:04:55 -04:00
GenomeAnalysisEngineUnitTest.java	Allow processing of empty intervals, but warn user when this case is encountered.	2011-10-28 12:12:14 -04:00
WalkerManagerUnitTest.java	Moving more tools over to use the standard VC arg collection. Also, while I'm in there, I removed all of the empty references to @Requires given that it's no longer relevant.	2011-08-10 12:20:18 -04:00