Guillermo del Angel 972d6531b6 Corner case fix for indel GL computation: sometimes (depending on surrounding context) reads which are not informative of two candidate haplotypes end up having marginally higher likelihoods with one haplotype as opposed to another, depending on uncertainty on alignments in surrounding regions. So, a sample whose GL is -0.0001,-0.0005,-0.001 may have its genotype set to 1/1 due to this statistical noise. We already have a tolerance comparing max(gl)-min(gl) to avoid genotyping, so this tolerance is now increased from 0.001 to 0.1 (equivalent to 1 PL unit) to avoid genotyping a sample if all PLs are within this threshold. Changed 2 integration test md5s that hit this case.		2012-04-26 10:15:26 -04:00
..
config	Reorganized the codebase beneath top-level public and private directories,	2011-06-28 06:55:19 -04:00
src	Corner case fix for indel GL computation: sometimes (depending on surrounding context) reads which are not informative of two candidate haplotypes end up having marginally higher likelihoods with one haplotype as opposed to another, depending on uncertainty on alignments in surrounding regions. So, a sample whose GL is -0.0001,-0.0005,-0.001 may have its genotype set to 1/1 due to this statistical noise. We already have a tolerance comparing max(gl)-min(gl) to avoid genotyping, so this tolerance is now increased from 0.001 to 0.1 (equivalent to 1 PL unit) to avoid genotyping a sample if all PLs are within this threshold. Changed 2 integration test md5s that hit this case.	2012-04-26 10:15:26 -04:00
test	Corner case fix for indel GL computation: sometimes (depending on surrounding context) reads which are not informative of two candidate haplotypes end up having marginally higher likelihoods with one haplotype as opposed to another, depending on uncertainty on alignments in surrounding regions. So, a sample whose GL is -0.0001,-0.0005,-0.001 may have its genotype set to 1/1 due to this statistical noise. We already have a tolerance comparing max(gl)-min(gl) to avoid genotyping, so this tolerance is now increased from 0.001 to 0.1 (equivalent to 1 PL unit) to avoid genotyping a sample if all PLs are within this threshold. Changed 2 integration test md5s that hit this case.	2012-04-26 10:15:26 -04:00