HPRC Pangenome Resources

This repo describes pangenomes produced by the Human Pangenome Reference Consortium from year 2 data. For information about data reuse and publishing with HPRC data please see the HPRC's Data Use Protocol.

Note: The pangenomes and resultant files referred to in this repo have not been fully QC'd, are not published, and may have known issues.

Background Information

Previous Version

For HPRC v1, click here.

Preprint

HPRC2: A human pangenome reference with near-complete coverage of common genetic variation

Graph Creation Strategies

Graphs are available from three different strategies summarized in the table (and relevant sections) below:

	Minigraph	Minigraph-Cactus	IMPG/PGGB
sequence comparison	reference-based, progressive	reference-based, progressive	symmetric, all-vs-all
resolution	SV only	base-level (via abPOA)	base-level (via seqwish/POA)
scope	full assemblies	Non-centromeric	full assemblies
cyclic paths	no	non-reference	all
short read mapping	untested	yes (fast)	untested
long read mapping	yes (fastest)	yes	yes (slowest)
Assembly mapping	yes (direct)	untested	yes (via injection)

Index files listing file locations for download with the AWS CLI can be found in the indexes folder of this repository. Alternatively, tables are listed below in each graph creation strategy's section. Note that the index files list the file locations with s3:// uris -- as opposed to http:// urls as found in the tables.

Assembly Inputs

Information about the source assemblies can be found in the HPRC Assembly GitHub repository. Of the 232 samples assembled (464 assemblies) in year 2, all but one were included in graph construction (HG00272 was held out due to an apparent large-scale misassembly in chrX). GRCh38 and CHM13 were added to make the total number of haplotypes included 464.

Graphs

Minigraph

Minigraph (cite) is a generalization of minimap2 (very fast) which builds the graph with iterative construction. Minigraph aligns with approximate locations and can be used to call structural variants (>50nt). Graphs were built with GRCh38, CHM13v2.0, and GRCh37 used as reference sequences. Note, these results were produced as part of Minigraph-Cactus (see below).

Description	GRCh38 Graph	CHM13 Graph	GRCh37 Graph
Graph	gfa	gfa	gfa

Minigraph-Cactus

Minigraph-Cactus (cite) adds base-level alignment to minigraph graphs.

The graphs below are v2.1. An earlier v2.0 version (built before the mitochondria-alignment and per-chromosome minigraph-ordering improvements) is documented here.

All steps to reproduce the graphs can be found here

Description	GRCh38 Graph	CHM13 Graph	GRCh37 Graph
Graph	gfa     gbz	gfa     gbz	gfa     gbz
Full (Unclipped) Graph	gfa     gbz	gfa     gbz	gfa     gbz
Chromosome Graphs	chroms	chroms	chroms
Multiple Alignment	HAL     MAF     MAF index	HAL     MAF     MAF index	HAL
Multiple Alignment (Duplications removed)	MAF     MAF index	MAF     MAF index	N/A
vcfwaved VCF	VCF     VCF index	VCF     VCF index	VCF     VCF index
VCF	VCF     VCF index	VCF     VCF index	VCF     VCF index
Raw VCF	VCF     VCF index	VCF     VCF index	VCF     VCF index
L95 VCF	VCF     VCF index	VCF     VCF index	N/A
PanGenie VCFs	biallelic     biallelic index     input     input index	biallelic     biallelic index     input     input index	N/A
VG Indexes	gbz     hapl     snarls	gbz     hapl     snarls	gbz     hapl     snarls
GBZ-Base Index	gbz.db	gbz.db	gbz.db
AF-Filtered VG Indexes	gbz     dist     min     zipcodes     snarls	gbz     dist     min     zipcodes     snarls	gbz     dist     min     zipcodes     snarls
Reference Gaps	bed	bed	N/A
Excluded Regions	bed     full bed     af bed	bed     full bed     af bed	bed     full bed     af bed
gRef Coordinates	gref	gref	N/A
All Files	files	files	files

The graphs are available in gfa format alongside other graph and index files. Information about the associated file formats can be found:

• mc output overview
• vg wiki
• gbz format (cite)
• hapl format: allows vg giraffe to infer a personalized pangenome as alternative to AF-filtering.
• maf: the HAL multiple alignment is also exported to indexed MAF (.maf.gz + .tai), which can be read and sliced with taffy.
• gbz-base: a database (.gbz.db) providing efficient random access to a GBZ graph without loading it into memory.

VCF Decomposition

Three VCFs are provided for each graph, from least to most processed:

• Raw VCF (.raw.vcf.gz): a site for each bubble in the graph. Nested bubbles result in overlapping sites, with the nesting relationships denoted by the PS (parent snarl), LV (level) and AT (allele traversal) tags, which need to be taken into account when interpreting the VCF.
• VCF (.vcf.gz): nested bubbles have been "popped" with vcfbub (removing alleles larger than 100kb), but the alt alleles have not been realigned.
• vcfwaved VCF (.wave.vcf.gz): as above, with each alt allele additionally realigned to the reference using vcfwave plus further normalization.

Unlike the v1.1 graphs (which used vcfbub and vcfwave only), the v2.1 decomposition applies additional normalization, built into Minigraph-Cactus via the --vcfwave option:

vcfbub -l 0 -a 100000
bcftools annotate -x INFO/AT
bcftools norm -m -any
vcfwave -I 1000
bcftools norm -f <ref.fa>
sort -k1,1d -k2,2n -s
merge_duplicates.py   # https://github.com/Han-Cao/collapse-bubble
bcftools norm -m +any
vcffixup

These extra steps (relative to v1.1) 1) pass biallelic input to vcfwave, which has had issues with multiallelic sites, 2) left-shift variants after vcfwave, and 3) merge the left-shifted variants back into multiallelic sites.

These steps are also described in the Minigraph-Cactus VCF documentation.

Evaluation Graphs

Versions of the CHM13 and GRCh38-based graphs are available here with HG002, HG005, and NA19240 excluded. Use these if, for example, you are running Genome-In-A-Bottle benchmarks.

Reference Assemblies

Slightly different reference assemblies were used in different graphs in order to handle the fact that CHM13v2.0 has chrY from HG002, and that GRCh38 unplaced contigs aren't wanted except when GRCh38 is the reference.

• hprc-v2.1-mc-chm13 : CHM13v2.0, GRCh38 (chromosomes only)
• hprc-v2.1-mc-grch38 : CHM13v2.0 (without chrY), GRCh38
• hprc-v2.1-mc-grch37 : CHM13v2.0 (without chrY), GRCh38 (chromosomes only), hs37d5
• hprc-v2.1-mc-chm13-eval : CHM13v2.0 (but with chrY from HG03017 instead of HG002), GRCh38 (chromosomes only)
• hprc-v2.1-mc-grch38-eval : CHM13v2.0 (without chrY), GRCh38

Filtered Graphs

The "AF-Filtered VG indexes" above were created by dropping nodes and edges supported by fewer than 10% of haplotypes, and give the best performance for Giraffe and are what have been used in the various papers to date. Note that giraffe requires only the .gbz, .dist and .min indexes.

PanGenie VCFs

The PanGenie VCFs (GRCh38 and CHM13 only) were normalized using PanGenie's pipeline for use with PanGenie:

• The .pgbi.vcf.gz (biallelic) files are an alternative to the .wave.vcf.gz files above. Rather than realigning large bubbles, this approach uses the paths in the graph to split them up.
• The .pgin.vcf.gz (input) files are the PanGenie input VCFs and can be used to genotype other samples with PanGenie.

Reference Gaps

Regions unaligned in the reference (predominantly satellite sequences) are provided in the refgaps BED file.

Excluded Regions

The Minigraph-Cactus pipeline clips out contigs (or pieces of contigs) that don't align confidently into the graph. Per-sample BED files of these removed intervals are bundled in the .missing.tar.gz archives, one archive per graph variant: the default graph (.missing.tar.gz), the full graph (.full.missing.tar.gz, which still excludes contigs that couldn't be confidently assigned to a reference chromosome), and the AF-filtered graph (.d46.missing.tar.gz).

gRef Coordinates and L95 VCFs

These outputs are still experimental.

• gRef Coordinates: pangenome graph-reference (gRef) coordinates. These include the base reference (CHM13 or GRCh38) along with contigs covering large insertions in the graph and enable description of variants nested within SV insertions. See the Minigraph-Cactus documentation for more information.
• L95 VCFs (.L95.vcf.gz): decomposed VCFs with SV alleles that are more than 95% similar merged together.

IMPG/PGGB

PGGB (the PanGenome Graph Builder) constructs a graph from an all-vs-all alignment of the input assemblies, giving base-level resolution with no reference bias. In this release, that same all-vs-all alignment is also provided through IMPG (implicit pangenome), which represents the pangenome as a set of assembly-vs-assembly alignments that can be projected and queried on the fly. The IMPG alignments, the PGGB graphs, and VCFs called from those graphs are provided below.

Graphs (PGGB)

Graphs were built with pggb -p 98 -k 311 and are provided as whole-genome, per-chromosome, and per-partition (community) GFAs. All files are zstd-compressed (.zst).

Description	Location
Whole-genome graph	gfa
Per-chromosome graphs	gfas
Partition graphs	gfas
All Files	files

VCFs (PGGB)

Variants decomposed from the PGGB graphs, relative to GRCh38. These are bgzipped VCFs with no tabix (.tbi) index.

Description	Location
Whole-genome VCF	vcf
Per-chromosome VCF	vcf

Alignments (IMPG)

Alignments are provided in PAF format and in IMPG's TPA format (one file per haplotype, 466 total).

Description	Location
All-vs-all alignment	paf
All haplotypes vs GRCh38	paf
All haplotypes vs CHM13	paf
Per-target alignments (all-vs-1)	pafs
Per-haplotype alignments (TPA)	tpas