Statistical Genomics Foundry
Paper
domain/pangenomics

Liao et al. 2023 — A draft human pangenome reference

1. Citation

Liao W-W, Asri M, Ebler J, et al. (Human Pangenome Reference Consortium). 2023. “A draft human pangenome reference.” Nature 617(7960):312–324. DOI: 10.1038/s41586-023-05896-x. Open access: PMC10172123 (https://pmc.ncbi.nlm.nih.gov/articles/PMC10172123/).

2. Access note

Full text read via PMC (PMC10172123). Extraction done against the PMC HTML. I did not open supplementary/extended-data files; all numbers below are from the main-text body as served by PMC. License: Creative Commons Attribution 4.0 International (CC BY 4.0) — page states the article “is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format.” Quoting mode: verbatim load-bearing quotes permitted (used in §7).

3. Thesis (1 sentence)

The HPRC presents a first-draft human pangenome reference built from 47 phased diploid assemblies of genetically diverse individuals that captures known and novel variation and, used as a graph reference, improves short-read variant discovery over the single linear GRCh38 reference.

4. Problem & context

The single linear reference (GRCh38) is one mostly-European mosaic that induces reference bias and misses sequence and structural variation present in other populations. The goal is a graph-based pangenome that represents many haplotypes simultaneously so that short-read analysis can map and genotype against diversity rather than against one arbitrary path. This is the “first release” of a consortium effort aiming at 350 individuals / 700 haplotypes.

5. Method / approach

  • Assemblies: 47 fully phased, diploid assemblies. Samples selected to represent global genetic diversity; 29 of the HPRC samples drawn from 1000 Genomes Project (1KG) lymphoblastoid cell lines, spanning African, East Asian, South Asian, European, and American ancestry groups.
  • Assembly quality: average QV 53.57 (≈1 base error per 227,509 bases); haploid assemblies with an X chromosome averaged 3.04 Gb total length (99.3% the length of T2T-CHM13); average haplotype switch-error rate 0.67%, Hamming error rate 0.79%. Assemblies cover >99% of expected sequence and are >99% accurate at structural and base-pair levels.
  • Graph construction: three graph-construction methods applied — Minigraph, Minigraph-Cactus (MC), and PanGenome Graph Builder (PGGB).
  • Anchoring references: both GRCh38 and T2T-CHM13 are incorporated within the pangenomes.
  • Downstream evaluation: short-read data mapped/genotyped against the draft pangenome (MC graph) and compared to GRCh38-based workflows for small-variant and SV discovery.

6. Key claims / findings (atomic)

  • The pangenome contains 47 phased, diploid assemblies (= 94 haplotypes) from a genetically diverse cohort.
  • Assemblies cover >99% of expected sequence per genome and are >99% accurate at structural and base-pair levels.
  • Adds 119 million bp of euchromatic polymorphic sequence and 1,115 gene duplications relative to GRCh38; roughly 90 million of the added bp derive from structural variation.
  • The MC graph contains 22 million small variants and 67,000 SVs.
  • Using the draft pangenome for short-read analysis reduced small-variant discovery errors by 34% and increased SVs detected per haplotype by 104% vs GRCh38-based workflows.
  • In “easy” unique regions (75.42% of the autosomal genome), samples showed mean 99.64% recall and 99.64% precision for small variants in the MC graph.
  • Variant-calling performance was lower in highly repetitive regions (3.87% of the autosomal genome) and lower for SVs than small variants.
  • On average ~136 Mb (4.4%) of T2T-CHM13 was not covered by any alignment.
  • PGGB vs Minigraph/MC differ in topology at copy-number-polymorphic loci (see §7 q3).

7. Load-bearing statements — VERBATIM (CC BY 4.0)

  1. (Abstract) “The pangenome contains 47 phased, diploid assemblies from a cohort of genetically diverse individuals. These assemblies cover more than 99% of the expected sequence in each genome and are more than 99% accurate at the structural and base pair levels.”
  2. (Abstract) “We also add 119 million base pairs of euchromatic polymorphic sequences and 1,115 gene duplications relative to the existing reference GRCh38. Roughly 90 million of the additional base pairs are derived from structural variation.”
  3. (Results) “Using our draft pangenome to analyse short-read data reduced small variant discovery errors by 34% and increased the number of structural variants detected per haplotype by 104% compared with GRCh38-based workflows.”
  4. (Results — repeat-region limitation) “Variant calling performance was lower in highly repetitive genome regions (3.87% of autosomal genome; Fig. 4a,b), for which more work will be required to achieve high-quality variant maps.”
  5. (Results — builder topology disagreement) “PGGB tends to collapse copy-number polymorphic loci like segmental duplications and VNTRs into a single copy through which haplotypes loop, while Minigraph and MC do not.” And: “In general, in the PGGB graph, many SV hotspots, including the centromeres, were transitively collapsed into loops through a subgraph representing a single repeat copy.”

8. Stated scope, assumptions, limitations (source’s own caveats)

  • Explicitly a draft / first release — “a draft human pangenome, the first release from the Human Pangenome Reference Consortium (HPRC),” aiming to “better capture global genomic diversity across the 700 haplotypes of 350 individuals.” Current 47 assemblies are a starting point, not the target.
  • Highly repetitive regions (3.87% of autosomal genome) underperform for variant calling and need more work.
  • SV performance lower than small-variant performance.
  • ~136 Mb (4.4%) of T2T-CHM13 was not covered by any alignment — “systematically unassembled or cannot be reliably aligned.”
  • Segmental-duplication assembly errors correlate with longer, more-identical SDs (median 288 kb vs 96.3 kb; 98.9% vs 97.1% identity).

9. Failure modes / invalidity patterns (referee-relevant)

  • Graph genotyping/variant calling degrades in highly repetitive regions. Verbatim: “Variant calling performance was lower in highly repetitive genome regions (3.87% of autosomal genome…), for which more work will be required to achieve high-quality variant maps.” So high-confidence pangenome-based genotype claims are limited to the “easy,” relatively unique ~75% of the autosome; centromeres, segmental duplications, and VNTRs are the weak zones.
  • Graph builders disagree on topology at repeats. PGGB collapses copy-number-polymorphic loci (SDs, VNTRs, centromeres) into single-copy loops that haplotypes traverse; Minigraph and MC do not. So the same input yields structurally different graphs at exactly the loci where genotyping is already least reliable — the graph representation itself is method-dependent there.
  • SV genotyping less reliable than small-variant genotyping (performance “lower for SVs than for small variants”).
  • Alignment gaps: ~4.4% of T2T-CHM13 uncovered → some regions systematically unassembled/unalignable.

10. What the source does NOT address (confident silences)

  • PanGenie is not named in the main-text content read; the paper’s short-read evaluation uses the MC graph pipeline, not PanGenie by name. (Guidance q2 named PanGenie only as an example; the repeat-region limitation is stated for the paper’s own graph genotyping, not attributed to PanGenie.)
  • No explicit allele-balance-at-heterozygous-sites metric for reference-bias reduction was found in the read content; reference-bias improvement is quantified as the 34% error reduction / 104% SV increase, not as allele-fraction balance.

11. Open questions / ambiguities

  • Exact per-sample variant counts vs the graph-aggregate counts (22M small variants, 67,000 SVs are graph-level) not separated in the read content.
  • How much of the builder-topology disagreement propagates into differing genotype calls is described qualitatively, not quantified in the read content.

12. Guidance answers

q1 — haplotype/sample count and variants resolved. 47 phased diploid assemblies (94 haplotypes). Graph-level variant content in the MC graph: 22 million small variants and 67,000 SVs. Adds 119 Mb euchromatic polymorphic sequence and 1,115 gene duplications vs GRCh38 (~90 Mb from SVs). NOTE: the claim-under-test’s “90 haplotypes, ~6.4M variants” does NOT match — the paper says 94 haplotypes and 22M small variants + 67,000 SVs in the MC graph.

q2 — graph-genotyping limitation in repetitive regions (referee-relevant, verbatim). “Variant calling performance was lower in highly repetitive genome regions (3.87% of autosomal genome; Fig. 4a,b), for which more work will be required to achieve high-quality variant maps.” (Contrast: “easy” unique regions = 75.42% of the autosome at 99.64% recall/precision.) The paper does NOT name PanGenie; this limitation is stated for its own graph-based variant calling.

q3 — graph builders and topology disagreement (verbatim). Three builders: Minigraph, Minigraph-Cactus (MC), and PanGenome Graph Builder (PGGB). They disagree on topology: “PGGB tends to collapse copy-number polymorphic loci like segmental duplications and VNTRs into a single copy through which haplotypes loop, while Minigraph and MC do not.” And centromeres/SV hotspots are “transitively collapsed into loops” in PGGB.

q4 — reference-bias reduction quantified. Yes: “reduced small variant discovery errors by 34% and increased the number of structural variants detected per haplotype by 104% compared with GRCh38-based workflows.” (No allele-balance metric found in read content.)

q5 — anchoring reference(s) (verbatim). Both “GRCh38 and T2T-CHM13 references within the pangenomes.” Assembly length compared to T2T-CHM13 (haploid X-bearing assemblies = 99.3% its length).