Help

Population genetics glossary

The terms of ancient DNA ancestry modeling, as the help guides and Nexju use them. Each links to the guide that explains it in full.

1240K: A set of about 1.2 million SNPs that most ancient DNA studies capture and genotype. The AADR’s 1240K release covers most ancient genomes well. Read the guide
AADR (Allen Ancient DNA Resource): The curated dataset of published ancient and present-day human genomes, with one annotation row per genome: its date, site, group label and coverage. Most qpAdm studies use it. Read the guide
Admixture: The mixing of two or more previously separated populations. A population formed this way carries ancestry from each source in some proportion. Read the guide
allsnps: A qpAdm option. When it is off, every statistic uses only the SNPs all populations of the model share. When it is on, each f4-statistic uses every SNP its own four populations share, which keeps more data for low-coverage genomes. Read the guide
Block jackknife: The way f-statistic standard errors are estimated: the genome is split into blocks of about 5 centimorgans, the statistic is recomputed with each block left out, and the spread of those values gives the standard error. Read the guide
BP (before present): Years before 1950. A genome dated 5,000 BP lived about 3050 BCE. Read the guide
D-statistic (ABBA-BABA test): A normalised f4-statistic that tests for gene flow from the counts of two site patterns in four populations. A D near 0 fits a tree; a D far from 0 shows gene flow. Read the guide
DATES: A method that estimates when two ancestries mixed from how the correlation of ancestry along the genome decays with distance. It works on one genome. Read the guide
EIGENSTRAT format: The three-file genotype format of ADMIXTOOLS and EIGENSOFT: .geno (genotypes), .snp (SNP list) and .ind (genome, sex and group label). The AADR is published in it. Read the guide
f2-statistic: The expected squared difference in allele frequency between two populations: a measure of the genetic drift that separates them. Read the guide
f3-statistic: f3(C; A, B) measures the drift C shares with A and B. A significantly negative value shows that C is a mixture of populations related to A and B. Read the guide
f4-statistic: f4(A, B; C, D) compares the drift of the pairs (A, B) and (C, D). A value of 0 fits a tree in which the two pairs split apart; a value far from 0 shows gene flow. qpAdm and qpWave are built from f4-statistics. Read the guide
f4-ratio: The ratio of two f4-statistics. For a known tree with one admixture event it estimates the mixture proportion. Read the guide
Feasible model: A qpAdm model whose weights all lie between 0 and 1. A model with a negative weight is infeasible even when its p-value passes. Read the guide
FST: The share of genetic variation that lies between populations rather than within them. Nexju uses the Hudson estimator. Read the guide
G25 coordinates: A genome’s position on 25 principal components, used for fast distance and mixture fits such as nMonte. These fits give no test of whether a model holds. Read the guide
Genetic drift: Random change in allele frequencies from one generation to the next. Populations that share history share drift, which f-statistics measure. Read the guide
Human Origins: A SNP array of about 600,000 SNPs typed in many present-day populations. The AADR’s Human Origins release suits PCA and present-day reference groups. Read the guide
Left populations: In qpAdm, the target and its candidate sources; in qpWave, the populations whose relation to the right populations is tested. Read the guide
Nested model: A qpAdm model with one source left out. If a nested model still fits, the source it leaves out may not be needed. Read the guide
p-value (qpAdm): The probability of a misfit at least as large as the observed one if the model is true. Most studies accept a model at p ≥ 0.01; a higher p-value does not make a model better. Read the guide
PCA (principal component analysis): A method that places genomes on axes of genetic variation. In ancient DNA work the axes are computed from present-day genomes and the ancient genomes are projected onto them. Read the guide
Pseudohaploid: A genotype made by taking one read at random at each SNP, which gives one allele instead of two. Most low-coverage ancient genomes in the AADR are stored this way. Read the guide
qpAdm: A test of whether a target population is a mixture of chosen sources, relative to a set of right populations. It reports a p-value for the fit and the weight of each source with its standard error. Read the guide
qpGraph: A method that fits an admixture graph, a population tree with mixture edges, to f-statistics and reports the worst-fitting statistic as a Z-score. Read the guide
qpWave: A rank test of how many streams of ancestry relate the left populations to the right populations. Rank 0 means the left populations are cladal relative to the right ones. Read the guide
Right populations (outgroups): The reference populations of a qpAdm or qpWave model. They must be related to the sources in different ways, and they must outnumber the sources. Read the guide
Rotation: A qpAdm strategy that tests many models from one pool of candidates: each model takes some candidates as sources and puts the others among the right populations. Read the guide
SNP: Single-nucleotide polymorphism: a position in the genome where people carry different bases. Ancient DNA analyses compare genomes at a fixed set of SNPs.
Source population: A population that a qpAdm model proposes as one ancestor of the target. It stands in for the real ancestral group, which was rarely sampled. Read the guide
Standard error: The uncertainty of an estimate such as a qpAdm weight, from the block jackknife. A weight of 0.40 with a standard error of 0.05 is 0.30 to 0.50 at about two standard errors. Read the guide
Target: The population or genome that a qpAdm model, f-statistic or DATES run describes. In Nexju it can be an AADR population, a custom group of genomes or your own uploaded DNA. Read the guide
Weights (admixture proportions): The share of the target’s ancestry that a qpAdm model assigns to each source. They sum to 1. Read the guide
Z-score: An estimate divided by its standard error. In f-statistics, |Z| ≥ 3 is the usual threshold for a result that differs from 0. Read the guide