granite is a collection of software to work with genomic variants. The suite provides inheritance mode callers and utilities to filter and refine variants called by other methods in VCF format.
granite library can also be used through an API to manipulate files in VCF format.
For more details, see granite documentation.
A ready-to-use docker image is available to download.
docker pull b3rse/granite:<version>
Additional software needs to be available in the environment:
To install the program from source, run the following commands:
git clone https://github.com/dbmi-bgm/granite
cd granite
make configure
make update
make build
To install the program with pip:
pip install granite-suite
The program is compatible with standard BED, BAM and VCF formats (VCFv4.x).
RCK is a tabular format that allows to efficiently store counts by strand (ForWard-ReVerse) for reads that support REFerence allele, ALTernate alleles, INSertions or DELetions at CHRomosome and POSition. RCK files can be further compressed with bgzip and indexed with tabix for storage, portability and faster random access. 1-based.
Tabular format structure:
#CHR POS COVERAGE REF_FW REF_RV ALT_FW ALT_RV INS_FW INS_RV DEL_FW DEL_REV
13 1 23 0 0 11 12 0 0 0 0
13 2 35 18 15 1 1 0 0 0 0
Commands to compress and index files:
bgzip PATH/TO/FILE
tabix -b 2 -s 1 -e 0 -c "#" PATH/TO/FILE.gz
BIG is a hdf5-based binary format that stores boolean values for each genomic position as bit arrays. Each position is represented in three complementary arrays that account for SNVs (Single-Nucleotide Variants), insertions and deletions respectively. 1-based.
hdf5 format structure:
e.g.
chr1_snv: array(bool)
chr1_ins: array(bool)
chr1_del: array(bool)
chr2_snv: array(bool)
...
...
chrM_del: array(bool)
note: hdf5 keys are built as the chromosome name based on reference (e.g. chr1) plus the suffix specifying whether the array represents SNVs (_snv), insertions (_ins) or deletions (_del).
When the program requires pedigree information, the expected format is as follow:
[
{
"individual": "NA12877",
"sample_name": "NA12877_sample",
"gender": "M",
"parents": []
},
{
"individual": "NA12878",
"sample_name": "NA12878_sample",
"gender": "F",
"parents": []
},
{
"individual": "NA12879",
"sample_name": "NA12879_sample",
"gender": "F",
"parents": ["NA12878", "NA12877"]
}
]
where individual is the unique identifier for member inside the pedigree, sample_name is the corresponding sample ID in VCF file, and parents is the list of unique identifiers for member parents if any.
granite <command> ...
positional arguments:
<command>
novoCaller Bayesian de novo variant caller
comHet compound heterozygous variant caller
mpileupCounts
samtools wrapper to calculate reads statistics for pileup at
each position
blackList utility to blacklist and filter out variants from input VCF
file based on positions set in BIG format file and/or
population allele frequency
whiteList utility to whitelist and select a subset of variants from
input VCF file based on specified annotations and positions
cleanVCF utility to clean INFO field of input VCF file
geneList utility to clean VEP annotations of input VCF file using a
list of genes
toBig utility that converts counts from bgzip and tabix indexed RCK
format into BIG format. Positions are "called" by reads
counts or allelic balance for single or multiple files (joint
calls) in specified regions
rckTar utility to create a tar archive from bgzip and tabix indexed
RCK files. Creates an index file for the archive
qcVCF utility to create a report of different metrics calculated
for input VCF file
validateVCF utility to calculate error models for input VCF file using
pedigree information
novoCaller is a Bayesian calling algorithm for de novo mutations. The model uses read-level information both in pedigree (trio) and unrelated samples to rank and assign a probabilty to each call. The software represents an updated and improved implementation of the original algorithm described in Mohanty et al. 2019.
warning: starting from version 0.1.12, novoCaller --triofiles expected order changed. Now PROBAND must be listed as first.
comHet is a calling algorithm for compound heterozygous mutations. The model uses genotype-level information in pedigree (trio) and VEP-based annotations to call possible compound heterozygous pairs. VEP annotations are used to assign variants to genes and transcripts, genotype information allows to refine calls based on inheritance mode. Calls are further flagged as "Phased" or "Unphased", where "Phased" means that genotype information supports in-trans inheritance for alternate alleles from parents.
blackList allows to filter-out variants from input VCF file based on positions set in BIG format file and/or provided population allele frequency.
whiteList allows to select and filter-in a subset of variants from input VCF file based on specified annotations and positions. The software can use provided VEP, ClinVar or SpliceAI annotations. Positions can be also specfied as a BED format file.
cleanVCF allows to clean INFO field of input VCF file. The software can remove a list of TAG from INFO field, or can be used to clean VEP annotations.
geneList allows to clean VEP annotations by applyng a list of genes. The software removes all the transcripts that do not map to a gene on the list.
qcVCF produces a report in JSON format with different quality metrics calculated for input VCF file. Both single sample and family-based metrics are available.
mpileupCounts uses samtools to access input BAM and calculates statistics for reads pileup at each position in the specified region, returns counts in RCK format.
toBig converts counts from bgzip and tabix indexed RCK format into BIG format. Positions are "called" by read counts or allelic balance for single or multiple files (joint calls) in specified regions. Positions "called" are set to True (or 1) in BIG binary structure.
rckTar creates a tar archive from bgzip and tabix indexed RCK files. Creates an index file for the archive.
validateVCF allows to calculate error models for different inheritance modes for input VCF file using pedigree information.