github.com/vgteam/vg_wdl/vg_mapgaffe_call_sv_cram

Read mapping and SV genotyping using vg. This workflow uses the new and faster mapper (giraffe). It takes a CRAM file and indexes for a graph containing the structural variants to genotype. The XG, minimizer and distance graph indexes as required. Including the snarls index is optional but speeds up the computation. It outputs a VCF file. By default, the VCF contains SV calls in the context of the snarls in the graph (i.e. variant site/bubbles). This helps merging the calls from multiple samples that used the same input graph. However it contains additional flanking sequences in the REF/ALT fields so should be normalized (e.g. using 'bcftools norm -f REF.fa VCF.vcf') before comparing with other SV catalogs or annotating the variants. If you are sure you don't want this 'snarl calling' behavior, set SNARL_CALLING to false. Finally, if the graph was built using a VCF and you want to genotype SVs using that same VCF, pass it to VCF_FILE and switch off the SNARL_CALLING which is not useful in that case. The CRAM files are converted to FASTQ in chunks. This is used to parallelize the CRAM conversion jobs and the mapping jobs. It also makes them short enough that they can be run on cheaper instances (e.g. pre-emptible). The NB_CRAM_CHUNKS parameter controls the number of chunks (~15 recommended for 20x Illumina WGS). The MAX_CRAM_CHUNKS parameter can be used to down-sample the reads by using only chunks up to MAX_CRAM_CHUNKS. For example, with NB_CRAM_CHUNKS=15 and MAX_CRAM_CHUNKS=5 only the first 5 chunks will be used, leading to downsampling 1/3 of the reads. To allow for preemptible instance, increase the PREEMPTIBLE attempts number parameter, e.g. PREEMPTIBLE=3.