A pangenome graph holds many human genomes at once. Sequence that the genomes share collapses onto common paths, and the places where they differ show up as bubbles in the graph. This track shows the structural variants found in version 2.1 of the Human Pangenome Reference Consortium (HPRC) minigraph-cactus graph, which was built from haplotype-resolved PacBio HiFi assemblies of 233 samples. Only larger events are shown here: insertions and deletions of at least 50 bp. HPRC produces one variant file per reference path, so the events are measured against GRCh38 on hg38 and against T2T-CHM13 on hs1, and each assembly shows its own native callset.
On hg38 there are about 550,000 such alleles (roughly 422,000 insertions and 128,000 deletions). On hs1 there are about 541,000 (roughly 348,000 insertions and 193,000 deletions). The two sets are not lifted between assemblies; the counts differ because an insertion against one reference can be a deletion against the other.
Items are colored by SV type:
| Insertion (INS) | |
| Deletion (DEL) |
An insertion is drawn as a 1 bp anchor at the point where the extra sequence goes in. A deletion spans the stretch of reference that is missing. Each variant keeps its allele count, allele frequency, the number of samples with data, and the level it sits at in the graph's snarl tree. A snarl level of 0 is a top-level bubble; higher numbers are bubbles nested inside a parent bubble. All of these can be used as filters.
HPRC release 2 does not yet have a peer-reviewed paper. The graph was built with minigraph-cactus from haplotype-resolved PacBio HiFi assemblies of 233 samples, including T2T-CHM13 and the diverse 1000 Genomes Project panel, using GRCh38 as the reference path. Variants were called from the graph with vg deconstruct. HPRC keeps the sample list and assembly provenance in alignments_v2.0.csv.
We started from the per-reference files provided by the HPRC graph team, hprc-v2.1-mc-grch38.gref95.ro.vcf.gz for hg38 and hprc-v2.1-mc-chm13.gref95.ro.vcf.gz for hs1. These are the raw vg deconstruct output: each graph bubble is one multi-allelic record with its graph traversals attached, and there are no per-allele type or length fields. To turn a file into a track, we compared every alternate allele to the reference allele after trimming the sequence they share at each end. An allele was kept when the net length change was at least 50 bp, and labeled an insertion when the alternate is longer or a deletion when it is shorter. At this size no balanced, equal-length substitutions came up, and the files carry no inversion calls, so the track has only insertions and deletions. On hg38, 549,649 alleles were kept (40,678 at nested snarl levels); on hs1, 541,176 (70,200 nested), after removing byte-identical duplicate records. Because these files are not broken down into atomic indels, one bubble can appear as a single large allele rather than several small ones, so the counts are not comparable to a wave-decomposed callset. Allele counts, frequencies and sample counts come straight from the VCF.
The conversion script and autoSql schema are in makeDb/scripts/lrSv and the build steps are in the makeDoc at doc/hg38/lrSv.txt.
The data can be explored interactively in table format with the Table Browser or the Data Integrator, and read programmatically through our API, track=hprc2v21Sv. For automated download and analysis the variants are in a bigBed file on our download server, one per assembly: hg38 and hs1. You can pull out one region or the whole set with bigBedToBed, for example bigBedToBed http://hgdownload.soe.ucsc.edu/gbdb/hg38/lrSv/hprc2v21.bb -chrom=chr21 -start=0 -end=100000000 stdout.
Thanks to the Human Pangenome Reference Consortium for building and releasing the release-2 minigraph-cactus pangenome, and to Glenn Hickey for the v2.1 deconstructed VCF.
HPRC release 2 is not yet described in a peer-reviewed publication. The release announcement has background and data-access details: HPRC data release 2.