Which Fst should I use?

Hudson's Fst, as Bhatia et al. (2013) recommend, when comparing pairs of populations of different sizes; Weir and Cockerham's θ when you want one estimate over several populations at once. Patterson's is available for biallelic sites. All three report an average over sites as a ratio of sums, with a jackknife standard error.

Why is Hardy–Weinberg failing across my whole cohort?

Pooling differentiated populations produces a heterozygote deficit on its own (the Wahlund effect). Test within one population.

Do I need phased genotypes for LD?

No. Rogers–Huff r works from genotype counts directly, and haplotype r² and D′ are estimated by maximum likelihood from unphased genotypes.

Is anything uploaded?

No. Every calculation runs in the browser on the files in your project.

Sequencing

Population genetics: allele frequencies, Hardy–Weinberg, Fst, linkage disequilibrium

Read a cohort's genotypes population by population: allele frequencies, heterozygosity and exact Hardy–Weinberg tests per site, Fst between populations with standard errors, linkage disequilibrium between every pair of sites, and principal components that show how the samples are structured. It reads a VCF with genotypes and a sample table that groups the samples, and every number is checked against scikit-allel, bcftools or PLINK over the 1000 Genomes genotypes across APOL1.

Open Population genetics: allele frequencies, Hardy–Weinberg, Fst, linkage disequilibrium Open the Population genetics sample

Every 1000 Genomes phase 3 genotype across APOL1, 2,504 people in 26 populations, for allele frequencies, Hardy–Weinberg, Fst, LD and PCA

What it does

01Allele counts and frequencies, missingness, and heterozygous and homozygous calls per site and per sample, for all samples or one population
02Observed and expected heterozygosity and the inbreeding coefficient F per site
03The exact Hardy–Weinberg test with its excess-heterozygosity tail, per alternate allele
04Fst between populations: Weir and Cockerham over many at once, Hudson and Patterson for every pair, with block-jackknife standard errors
05Linkage disequilibrium between every pair of sites: Rogers–Huff r from genotypes, or haplotype r² and D′ by maximum likelihood
06LD pruning by sliding windows
07Principal components of genotypes with Patterson, standard or centring scaling, coloured by population

What it runs on

Faithful ports of scikit-allel, bcftools and PLINK 1.9, each checked against the real program over every site and sample of the 1000 Genomes phase 3 genotypes across APOL1. The panel runs on @ports/popgen.

Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.

Questions

Which Fst should I use?: Hudson's Fst, as Bhatia et al. (2013) recommend, when comparing pairs of populations of different sizes; Weir and Cockerham's θ when you want one estimate over several populations at once. Patterson's is available for biallelic sites. All three report an average over sites as a ratio of sums, with a jackknife standard error.
Why is Hardy–Weinberg failing across my whole cohort?: Pooling differentiated populations produces a heterozygote deficit on its own (the Wahlund effect). Test within one population.
Do I need phased genotypes for LD?: No. Rogers–Huff r works from genotype counts directly, and haplotype r² and D′ are estimated by maximum likelihood from unphased genotypes.
Is anything uploaded?: No. Every calculation runs in the browser on the files in your project.

Also searched as

allele frequency calculator · Hardy-Weinberg equilibrium test · HWE exact test for VCF · observed and expected heterozygosity · inbreeding coefficient F · Fst calculator · Weir and Cockerham Fst · Hudson Fst · linkage disequilibrium r2 and D prime · LD pruning · genotype PCA · population structure PCA · 1000 Genomes population analysis