Home

Awesome

DOI

Computationally phased T2T 1KGP panel

Download panel at this location

This repository contains CHM13v2-aligned 1000 Genomes Project (1KGP) variant data (Rhie et al 2022) that have been computationally phased using SHAPEIT5 (Hofmeister et al 2022). Phased panels (both unrelated 2504 member panels and full 3202 member panels) are available at the T2T/HPRC aws bucket.

A tar file containing the initial version of our phased T2T panel, along with another files containing phasing statistics has been uploaded to zenodo at this link. Please cite the zenodo repository if you find this panel to be useful in your work. A preprint is in the works, and once it is up we will have a proper citation for you.

Repository structure

Reference data sources

Switch Error Rates

Performance was measured by comparing the phased haplotypes of 39 samples shared between the Human Pangenome Reference Consortium (HPRC)'s draft human pangenome and the 1000 Genotypes dataset.

Method of phasing variantsSwitch Error Rate+/- 95th c.i.Ground Truth Source
All 3202 samples, including trios, with pedigree0.07%0.003%Trio concordance
2002 singletons or trio children, no pedigree1.31%0.03%Trio concordance
44 HPRC samples, using the 2504 unrelated samples panel (with HPRC samples and parents removed) as a reference panel2.29%0.10%HPRC concordance

Methods

Variant filters used in this analysis (And string used to implement filter with bcftools)

<br>

Note: the SHAPEIT5 UK Biobank phasing paper excludes alternative alleles with AAscore < 0.5. This is a statistic produced by GraphTyper, which was not used to produce this dataset. The closest equivelent is the VQSLOD produced by Haplotype Caller. GraphTyper's AA score is simply the likelihood of an alternative allele truly being present in the dataset, so a cutoff of 0.5 is equivelent to 50% odds. Log odds of 1:1 is 0, so the VQSLOD log odds equivelent would be to exclude sites with a VQSLOD score of less than a cutoff of 0.

<br>

Phasing

Phasing was performed with SHAPEIT v5.1.1, largely in accordance with the recommendations outlined in SHAPEIT5's online tutorial. For each chromosome, common variants were phased in one chunk. Rare variants were phased in chunks of 40 megabases.

Chromosome X PAR regions were phased separately from the rest of chromosome X. To phase the body of chromsome X, male samples were provided to SHAPEIT5 via the --halpoid option. The provided Ne value was reduced to 75% of the overall Ne, to account for the reduced population of haplotypes in this region of chrX. Phasing statistics were only calculated for female samples.

Evaluation of phasing accuracy

We rely on two different methods of phasing quality evaluation:

To perform these evaluations, we phase the data set four times:

We then calculate different sets of performance statistics using SHAPEIT5's switch tool.

Using family data as 'ground truth', we can:

Using the 39 1KGP samples present in the draft pangenome, we can use the HPRC's empirically phased haplotypes to: