Larrivhan/NeCast-1000G

NeCast 1000 Genomes analyses

NeCast's analyses of the 1000 Genomes Project: size histories of 25 populations under four map inputs, and the comparison of the low-coverage (phase 3) and 30× call sets. Everything between the public call sets and the published curves is here: the windows, the groups of individuals, the genotype slices and extractions, every NeCast input and estimate, the phlash fits of the call-set comparison, and the result tables.

Contents

path	what it holds	size
reference/	sample panel, accessibility masks, centromere and gap tables, download list with checksums, the scripts that fetched them	0.15 GB
windows/	the 51 windows of 30 Mb accessible sequence on chromosomes 1–18	0.80 GB
samples/	the populations' unrelated individuals and the groups of 32 used in each run	small
genotypes_30x/	per-window VCF slices of the 30× call set and the genotypes extracted from them	61.4 GB
populations_25/	the 25-population analysis under four map inputs, per chromosome, and its results	48.5 GB
low_coverage/	phase 3 slices and extractions, NeCast and phlash on both call sets, and the comparison	85.6 GB
figures/	the figure on this card	small
SHA256SUMS	a checksum for every file except this card	small

Data and windows

a, The 51 windows on GRCh38, alternately purple and light purple, with the number on each chromosome; grey blocks are centromeres. Chromosomes 19–22 hold no window. b, NeCast's estimates with both maps for YRI, CEU, GIH, CHB and PEL, read as an inverse coalescence rate (IICR). Each thin line is one window, averaged over the population's groups of 32, and the thick line is their mean. The values are in populations_25/results/curves_bundle4.json.

• Call set. The phased high-coverage (30×) call set (Byrska-Bishop et al. 2022), GRCh38. The original VCFs are not copied here; reference/DOWNLOADS.json gives their addresses, sizes and SHA-256, and reference/scripts/ the download script.
• Populations. 25 of the 26 populations. Each dataset holds 64 unrelated individuals, as a simulated training window holds 64 diploid genomes; ASW has 61 and was left out, since admitting relatives would inflate recent coalescence. NeCast reads disjoint groups of 32 of these individuals: two groups in 18 populations and one in seven, 43 analyses per window.
• Windows. 30 Mb of accessible sequence under the strict mask, within chromosome arms, with gaps of at most 1 Mb: 51 windows on chromosomes 1–18. Chromosomes 19–22 hold no such block. Accessible intervals are joined into one 30 Mb coordinate. windows/windows_chrN.json lists each window's intervals.
• Sites. Biallelic SNVs only; positions with more than one record dropped. This call set has no missing calls and leaves out variants seen once in its full sample. The reference allele is taken as ancestral.
• Rates. Mutation rate 1.29 × 10⁻⁸ per bp per generation. The recombination map is the deCODE 2019 average map and the mutation map the Roulette model; both are cut to each window's accessible intervals.

The 25-population analysis (populations_25/)

NeCast was run on the same rows with four map inputs:

folder	recombination	mutation
no_map	flat at the genome-wide mean	flat
mutation_map	flat at the genome-wide mean	Roulette, its shape rescaled to the window's mean rate
recombination_map	deCODE 2019	flat
both_maps	deCODE 2019	Roulette, rescaled as above

Each populations_25/chrN/<map input>/ holds rows/ (the windows in NeCast's input format, with REAL_ROWS.json naming the window, population, group and individuals of every row), inputs/ (summary statistics and PSMC+ posteriors), estimates/ (NeCast's predictions), and, for the two runs made directly on the extracted rows, range_checks/ (each row's inputs compared with the range of the training data) and tables/ (per-row predictions as rows.csv, rows.jsonl and predictions_log10_iicr.npz). Each dataset also holds a few simulated rows that are not part of the analysis; they come from decoy_sources/.

populations_25/results/ has the population curves and contrasts: curves_bundle4.json (every curve per population, window and group, and the per-interval 95% intervals), per_population4.tsv, per_population_decade4.tsv, per_window4.tsv and per_bin_ci4.tsv. A population's curve is the mean over its windows and groups; intervals are percentile bootstraps over windows, paired across map inputs (4,000 resamples). run_scripts/ has the drivers and collectors that produced them.

Low-coverage and 30× call sets (low_coverage/)

For YRI, CEU, CHB, GIH and PEL, the first group of 32 individuals and the 51 windows were extracted from the phase 3 call set (Auton et al. 2015; low coverage, mean depth 7.4×; GRCh38 release 20190312, biallelic SNVs), with the same filters, masks and maps.

• phase3_vcf_slices/ and phase3_extracted/: the phase 3 slices and extracted genotypes.
• necast_phase3/chrN/: NeCast with both maps on the phase 3 data; the 30× runs are those in populations_25/chrN/both_maps/.
• phlash/: phlash fitted per window on both call sets (tasks_p3.tsv, tasks_nygc.tsv).
• necast_checks/: two checks made before the 30× runs were reused for the comparison. A re-run of chromosome 16 reproduced the 30× run. NA18498, one of the YRI individuals that fill a dataset to 64 but are not in the analysed group, is absent from phase 3 and was replaced by NA18520 in both call sets; re-extracting chromosome 2 with this change left every YRI prediction unchanged (30x_extracted_fill/ holds that extraction).
• analysis/: the scripts, the singleton tally and the results (results/E2_summary.json, results/R6_necast_vs_phlash.json).

File formats

Windows in NeCast's input format use the layout of the training data (four windows per .npz, genotypes bit-packed); see the card of Larrivhan/NeCast-simulations and its reader tools/necast_windows.py. Predictions are log10 Ne on 64 intervals from 50 to 500,000 generations ago, evenly spaced in log time.

Paths in provenance files

Logs, plans and other provenance files were written on the machines that ran the jobs. Absolute paths in them use placeholders: $WORK and $DATA for the working directories, $SW for installed software and ~ for the home directory. Checksums recorded inside the files match the released files.

Licence and sources

The files here are released under CC BY 4.0. They derive from the 1000 Genomes Project, whose data are public: the 30× call set (Byrska-Bishop et al. 2022, Cell 185:3426) and phase 3 (1000 Genomes Project Consortium 2015, Nature 526:68). The maps are the deCODE 2019 map (Halldorsson et al. 2019, Science 363:eaau1043) and the Roulette mutation rates (Seplyarskiy et al. 2023, Nature Genetics). The full maps are not copied here: reference/EXCLUDED_THIRD_PARTY.tsv gives their checksums, and generator/ in Larrivhan/NeCast-simulations has the scripts that build them. phlash: Terhorst (2025), Nature Genetics 57:2570.

Citation

The NeCast paper is in preparation; its reference will be added here.