prescience-bio/pop-1kgp

1000 Genomes high-coverage GRCh38 population panel

A chromosome-sharded PLINK 2 representation of the 3,202-sample 1000 Genomes high-coverage GRCh38 callset. It contains 73,759,911 variant records across chr1-chr22, chrX, chrY, and chrMT.

The 75 published payloads occupy about 5.5 GiB. Exact source files, derivations, metadata corrections, and fidelity boundaries are recorded in data/artifact.yaml.

Data layout

Each chromosome is one matching PLINK prefix directly under data:

data/chr01.pgen.zst
data/chr01.pvar.zst
data/chr01.psam

The same layout applies to chr01-chr22, chrX, chrY, and chrMT.

• pgen.zst is the genotype matrix with an outer Zstandard layer. Decompress it to pgen before use.
• pvar.zst contains alleles, coordinates, identifiers, and retained site annotations. PLINK 2 reads it directly with the vzs modifier.
• psam contains the ordered sample, pedigree, sex, population, and superpopulation metadata.

The chr1-chr22/X sources are phased GT-only VCFs. The chrY and chrMT raw sources also contain auxiliary FORMAT fields; those fields are deliberately not published here. This panel preserves PLINK-retained genotype calls, missingness, supported phase, alleles, and site metadata. The original URLs remain in data/artifact.yaml for retrieval when those auxiliary fields matter.

Verify and use

Both aggregate checksum manifests cover exactly the 75 payloads:

scripts/verify-artifacts.sh
scripts/verify-artifacts-blake3.sh

The first command verifies structure plus SHA-256 and BLAKE3. It also accepts the --sha256, --blake3, and --all options.

Materialize only the chromosome needed for an analysis:

scripts/materialize-shard.sh chr01
plink2 --pfile work/materialized/chr01/chr01 vzs --freq

The materialized PGEN is disposable. Files under work are ignored by Git.

Export the fields retained by PLINK to an indexed VCF:

scripts/export-vcf.sh chr01 exported/chr01.vcf.gz

The default export location is ignored by Git. An exported VCF is semantically reconstructed and is not expected to be byte-identical to its source VCF.

Export individual whole-genome VCFs

The friendly exporter defaults to the tracked superpop-mf-10-vcf preset: one male and one female from each super-population, including NA12878 as the EUR female. This command exports indexed VCFs across the complete genome using the current default record mode, all:

scripts/export.py

Choose participants directly and set a destination:

scripts/export.py \
  --participants NA12878 HG00112 \
  --destination export/two-participants

Use a TSV whose first column contains participant IDs:

scripts/export.py \
  --participants-file my-samples.tsv \
  --destination export/my-panel

Select the output format and records to retain:

scripts/export.py \
  --preset superpop-mf-10-vcf \
  --format bcf \
  --records variants \
  --destination export/superpop-mf-10-bcf

Supported formats are vcf.gz (recommended compressed text plus Tabix index), vcf (plain text), and bcf (compact binary plus CSI index). Record modes are:

• variants: only genotypes carrying an alternate allele; most compact.
• called: every nonmissing genotype, including homozygous reference.
• all (default): every cohort site, including missing genotypes; this is very large.

Limit a test or specialized export to selected chromosomes:

scripts/export.py --participants NA12878 --chromosomes 21 22 X

The Python CLI uses only the standard library. PLINK 2, bcftools, and zstd do the native genomics work; check that they are available with:

scripts/export.py --check

List named presets or inspect every option with:

scripts/export.py --list-presets
scripts/export.py --help

The older preset wrapper remains available:

scripts/export-preset.sh superpop-mf-10-vcf

The exact, reviewable sample list is in presets/superpop-mf-10-vcf/samples.tsv. One indexed VCF per person is written to export/superpop-mf-10-vcf/. This combines chr1-chr22, chrX, chrY, and chrM. By default each VCF contains every cohort site. With --records variants, homozygous-reference and missing calls are omitted and unused alleles at multiallelic sites are trimmed. In every mode, cohort frequency and quality annotations are removed while the phased genotype and INFO fields needed to describe structural variants are retained. Generated data under export/ is ignored by Git, while the preset is tracked and repeatable.

Each output directory also contains a samples.tsv manifest. Its file_path values are relative to the manifest, making the directory portable as a unit.

The selected rare-disease trio (NA19650, father NA19649, mother NA19648) in export/rare-disease-trio/ now uses original NYGC single-sample calls with measured quality, rather than the GT-only phased PLINK exports. The directory name is retained. Download and compact these originals with:

scripts/download-trio.sh
python3 scripts/compact-trio-originals.py

The downloader requires aria2c and Python 3.9+, saves three gVCFs and their indexes under work/sources/rare-disease-trio/, and verifies the publisher's MD5 checksums. The converter requires bcftools with view -A support (tested with 1.24). It removes reference-only records and <NON_REF>, remaps AD/PL, verifies retained FORMAT values, and writes indexed VCFs of about 178–185 MiB each. It refuses to overwrite existing outputs. For a separate rebuild:

scripts/download-trio.sh /path/to/originals
python3 scripts/compact-trio-originals.py \
  --source /path/to/originals --destination work/trio-rebuilt

The already-verified originals are also stored on linux@192.168.1.99:/mnt/heighliner/data/rare-disease-trio/; local full-size copies have been removed. Retain the remote full gVCFs for family reference-call evidence. Compact VCFs are raw calls with FILTER=., not a claim that every call passes quality thresholds. A few source calls lack AD/DP; see quality-completeness.json and the directory's README.txt.

All pedigree files contain real 1000 Genomes IDs, parent-child relationships and sex. pedigree.fam and pedigree.psam retain unknown disease status. pedigree.talos.ped uses that real pedigree with synthetic affected/unaffected labels and HPO terms for testing. hpo_terms.json and hpo_terms.tsv describe those synthetic terms; they are not observed donor phenotypes. Historical Talos results and old exports have been removed. The new quality-bearing VCFs have not been run through Talos annotation and filtering.

For an independent PLINK-derived comparison, use a separate destination:

python3 scripts/export-rare-disease-trio.py --destination work/trio-plink

That exporter defaults to carried variants and cannot restore genome-wide read quality. Do not force it to overwrite the current quality-bearing trio.

To generate a different balanced 10-person manifest, then export it:

scripts/select-export-samples.sh
scripts/export-individual-vcfs.sh

To pin other people while preserving the one-per-sex/per-super-population balance, repeat --include (a second requested sample cannot occupy an already-filled slot):

scripts/select-export-samples.sh --no-default \
  --include HG00096 --include HG00097

For a custom set, place the desired IIDs in the first column of any tab-delimited file and pass it directly:

scripts/export-individual-vcfs.sh --samples my-samples.tsv

Per-chromosome intermediate VCFs are cached by participant set under work/, so switching final formats or record modes reuses the expensive PLINK filter. Use --force to rebuild them and --threads to control native workers.

Rebuild from upstream

Download the complete input inventory into the ignored source workspace:

scripts/download.sh

This downloads the 2022 phased chr1-chr22/X VCFs and indexes, the 2020 chrY VCF and index, pedigree and population metadata, the corrected PLINK PSAM, and the two upstream manifests. For chrMT it retrieves only chrM through the remote index of the approximately 20 GiB 2020 others VCF.

Build everything or a single chromosome:

scripts/build.sh
scripts/build.sh chr22

Both commands use work/sources by default. To use a different source directory:

scripts/download.sh /path/to/sources
SOURCE_DIR=/path/to/sources scripts/build.sh

The build verifies the corrected PSAM checksum, validates each requested shard against its source VCF, and refreshes data/checksums.sha256 and data/checksums.blake3. Use FORCE=1 to rebuild a completed shard, THREADS to set parallelism, and KEEP_WORK=1 to retain intermediates.

Provenance and toolchain

• catalog.tsv is the Hugging Face artifact catalog.
• data/artifact.yaml is the complete upstream and derivation record.
• tools.tsv pins the PLINK 2, bcftools/htslib, Zstandard, and BLAKE3 versions used for this build.