Releases: dnanexus-rnd/GLnexus
v1.2.0
- improved DeepVariant configurations
Adds tuned configuration presets DeepVariantWES
and DeepVariantWGS
(and changes the DeepVariant
preset to be equivalent to DeepVariantWGS
). #190 @tedyun
--more-PL
The --more-PL
flag causes all PL vectors in the output pVCF to be filled in. This provides compatibility with downstream tools that won't handle any missing values (e.g. Beagle gl= imputation). However, it's slower and inflates the pVCF output size, without adding much meaningful information.
- The PL for 0/0 entries are populated from reference bands.
- PL entries for genotypes including alleles not present in a given sample gVCF are populated from the sample PL entries corresponding to the gVCF symbolic allele (<*> or <NON_REF>)
- Any other entries that can't be filled accurately are set to 0 or 990 as appropriate, instead of the VCF missing value (.).
--trim-uncalled-alleles
When genotype revision is enabled, it's possible for an allele to be included in the output pVCF but not actually called in any of the sample GT genotypes. This is rare and typically associated with marginal-quality variants. The --trim-uncalled-alleles
flag (which sets genotyper_config::trim_uncalled_alleles
) removes these by postprocessing each output row using htslib's bcf_trim_alleles
routine.
glnexus_cli
docker image: quay.io/mlin/glnexus:v1.2.0
v1.1.12
- Reduce
unifier_config::max_alleles_per_site
to 32 for gatk and weCall configuration presets- to control PL vector size explosion in large cohorts
- alleles exceeding this limit are "kicked out" into monoallelic sites
- Record the full configuration YAML (as a one-liner) in the pVCF header
- Set Number=R for AD INFO in pVCF header (@brentp)
- Fix CI builds on external contributor PRs
- Bump dependency versions
glnexus_cli
docker image: quay.io/mlin/glnexus:v1.1.12
v1.2.0-pre.0
The new glnexus_cli --more-PL
flag causes all PL vectors in the output pVCF to be filled in.
- The PL for 0/0 entries are populated from reference bands.
- PL entries for genotypes including alleles not present in a given sample gVCF are populated from the sample PL entries corresponding to the gVCF symbolic allele (<*> or <NON_REF>)
- Any other entries that can't be filled accurately are set to 0 or 999 as appropriate, instead of the VCF missing value (.).
The new mode is slower and inflates the pVCF output size, often for limited useful information added, compared to the default mode which omits these cases. However, it can provide compatibility with downstream programs which require 100% of PL values filled in (e.g. Beagle gl= imputation).
v1.1.11
v1.1.10: refactor ingestion exceptions (#174)
apply the exceptions in unit tests and clean up issues that were lurking there