What is the best way to find intervals for parallelization of joint variant calling?
In general I can think of 3 ways:
- Split on poly-N regions in the reference genome. e.g. via picard ScatterIntervalsByNs
- Split on regions without mapped reads
- Split every 1Mbp, book-ended or overlapping
With improvements in sequencing quality it is becoming more difficult to find areas that are obviously safe to split the analysis on.
- Less and smaller poly-N regions
- More regions of the genome can have reads mapped (if using e.g. (long) reads of variable length)
This leads to the risk of not much parallelization (i.e. max per chromosome) or splitting right in the middle of a variant.
What is the best way to parallelize joint variant calling?
(e.g. GATK GenomicDBImport and GenotypeGVCFs on many GVCFs)?
Without losing variants or getting double variants?
•
written
by
William
1 answer
I wrote a tool named: FaidxSplitter http://lindenb.github.io/jvarkit/FaidxSplitter.html , I'm not sure if it fulfill your needs.
it takes as input:
- the indexed reference genome
- a BED of gaps where I don't want any calling
- a BED of genes/regions where I don't want any 'cut'.
- w the size of each region
- x the overlap of between to region
the output is a BED file with all the regions to be called.
I then call the variants for each bed and I concatenate the vcf with bcftools, removing the duplicates.
as you said, you could use ScatterIntervalsByNs, use the regions with sharing 0 coverage or sharing a too high coverage. For human, there are some blacklisted regions defined by ENCODE.
I prefer to use the fasta for this. BAM coverage analysis is not fully deterministic, it changes with the set of BAM files under analysis and is computational expensive if looking at many samples/bam files. Some non human model organism now have really high quality reference genomes without many (long) poly-N regions.
Log in to answer this question.
-
Find reference genome regions spanned by only mapping quality 0 reads in multiple WGS samples
written by William
-
Reduce poly-N regions longer than 200bp back to 200bp in a reference genome FASTA
written by William
-
Joint variant calling on DeepVariant GVCFs using GATK GenotypeGVCFs
written by Matin •
-
Why isn't it a common practice to call variant for every position in human genome?
written by optimistsso4co3 •
-
Calculate genome regions to exclude for Structural Variant calling
written by William
-
How to call denovo variants from multiple trios cohort?
written by DareDevil
-
Split reference genome into callable regions by splitting on NNNN stretches
written by William
-
best way to do joint genotyping
written by bitpir
-
Is it possible to use haplotypes instead of SNPs for association testing with phenotypes?
written by William
-
Split Bam Files By Region For Parallel Variant Calling
written by William
