Talk:Population structure

Talk

Discussion for: Population structure • View • Links • Related changes

Use this page for verification notes, disputes, and action items. Keep narrative content on the article.

Page scope

Population structure covers what population structure is, why populations come to differ, the methods that read structure from genes and how those methods can mislead, its effect on other analyses, the surveys of cannabis with their plants and markers, and how far those surveys can be compared. The values of diversity and the account of hemp and drug-type pools belong to Genetic diversity; marker systems to Genetic marker; the hierarchy of F statistics to Heterozygosity; estimates of migration to Gene flow; ancestry along chromosomes to Introgression; species concepts to Cannabis taxonomy; trees and domestication to Phylogenetics of Cannabis.

Data verification

Record source conflicts, figures that differ between studies and sources not used here. Keep narrative content on the article.

Discussion

Sources

Felsenstein, Theoretical Evolutionary Genetics (text of December 2019), and Davinack, Molecular Ecology & Evolution: An Introduction (2024): the definition, the Wahlund effect, F_ST after t generations of drift, the island and stepping-stone models.
Meirmans and Hedrick (2011), Jost (2008), Bhatia and colleagues (2013), Holsinger and Weir (2009), Whitlock and McCauley (1999): what F_ST measures and how far it can be compared.
Excoffier and colleagues (1992), Patterson and colleagues (2006), McVean (2009), Novembre and Stephens (2008), Pritchard and colleagues (2000), Alexander and colleagues (2009), Rousset (1997), Pickrell and Pritchard (2012): the methods.
Evanno and colleagues (2005), Janes and colleagues (2017), Puechmaille (2016), Kalinowski (2011), Lawson and colleagues (2018): choosing K and reading bar plots.
Serre and Pääbo (2004), Rosenberg and colleagues (2005), Schwartz and McKelvey (2009), Meirmans (2012): clusters and gradients.
Yu and colleagues (2006), Price and colleagues (2006), Mazet and colleagues (2016): effects on association mapping and on models of population size.
Hillig (2005), Gilmore and colleagues (2007), Piluzza and colleagues (2013), Sawler and colleagues (2015), Lynch and colleagues (2016), Soorni and colleagues (2017), Dufresnes and colleagues (2017), Zhang Q. and colleagues (2018), Schwabe and McGlaughlin (2019), Henry and colleagues (2020), Zhang J. and colleagues (2020), Pan and colleagues (2021), Grassa and colleagues (2021), Schwabe and colleagues (2021), Ren and colleagues (2021), Watts and colleagues (2021), Chen and colleagues (2022), Woods and colleagues (2023), Benkirane and colleagues (2024), Halpin-McCormick and colleagues (2024), Mostafaei Dehnavi and colleagues (2025), Aina and colleagues (2025), Lynch and colleagues (2025): the surveys of cannabis.

Where the sources disagree

Whether any SNP is fixed between hemp and drug type. Sawler and colleagues (2015) report no fixed difference among 14,031 SNPs, the highest F_ST being 0.87. Soorni and colleagues (2017), reanalysing those data, report 9 SNPs at 1 in hemp and 0 in drug-type plants and 92 the other way round.
Which label lies nearer hemp. Sawler and colleagues put hemp nearer strains reported as indica. Dufresnes and colleagues (2017) find most of the drug varieties closest to hemp to be of sativa ancestry and name two possible explanations, shared descent and crossing with indica and agricultural varieties.
Clusters or gradients. Serre and Pääbo (2004) find gradients when individuals are sampled evenly and see no reason to assume major discontinuities between continents. Rosenberg and colleagues (2005) accept the gradients and also find clusters. They hold that Serre and Pääbo chose study designs under which clustering was less pronounced. Both work on human data.
The measure of differentiation. Jost (2008) holds that G_ST and its relatives do not measure differentiation. Meirmans and Hedrick (2011) keep F_ST beside F′_ST or D. Bhatia and colleagues (2013) note a suggestion that D shares the same problems.
The first division. Sawler and colleagues and Dufresnes and colleagues find hemp against drug type at K = 2. Lynch and colleagues (2016) find broad-leaflet drug types against the rest at K = 2 with fastSTRUCTURE. Zhang J. and colleagues (2020) find a split mainly by Chinese or foreign origin at K = 2.
The number of migration events in the pangenome. The results of Lynch and colleagues (2025) state six gene-flow events in the TreeMix graph. The methods state five migration events as the number selected.
The life history of cannabis. Soorni and colleagues relate the high share of variance within populations to perennial dioecious plants. Welling and colleagues (2016) describe cannabis as an annual herb.
Gene flow from F_ST. Zhang J. and colleagues read Nm of 5.0974 and F_ST of 0.0468 as frequent exchange of genes among cannabis germplasm worldwide. Whitlock and McCauley (1999) hold that such a conversion is rarely accurate.

How far the evidence goes

The values on the article between hemp and drug type (0.156, 0.20, 0.229) come from different plants, SNP sets and calculations. Sawler and colleagues and Woods and colleagues use the estimator of Weir and Cockerham (1984). Grassa and colleagues filter their 367 accessions to reduce structure from family relationships.
F_ST calculated between clusters that a program has built from the same genotypes tests groups formed to differ. Aina and colleagues (2025) also keep only plants with more than 70 per cent membership in one cluster before calculating it.
The variance shares in the table on the article use different levels. Aina and colleagues give 71 per cent among all samples, and Benkirane and colleagues give 65 per cent within individuals. Neither is the same level as "within populations" in Soorni and colleagues.
Serre and Pääbo, Rosenberg and colleagues and Novembre and Stephens work on human data. The simulations cited model hierarchical fragmentation, dispersal among populations or neighbour mating. None models a crop moved by people and bred as clones.
The subgroup counts Zhang J. and colleagues list by region (73 and 83) do not sum to their totals of 99 and 100.
The values of F_ST after t generations on the article are calculated from the formula in Felsenstein: (1 − 1/100)¹⁰ = 0.904.
Pan and colleagues (2021) place cultivars from central China, Henan among them, in both groups. Schwabe and McGlaughlin (2019) draw their reported proportions from one online database, and all their loci depart from Hardy–Weinberg proportions.
Cannabis is an outcrossing annual, so it falls on both sides of the comparison in Nybom (2004).

Terms

K: the number of groups assumed by a clustering program.
ΔK: the rate of change in the likelihood between successive values of K, after Evanno and colleagues.
FLOCK: a clustering program used by Lynch and colleagues (2016).
Discriminant analysis of principal components: a method that finds the axes separating groups, used by Soorni and colleagues and Henry and colleagues.
Spatial analysis of molecular variance: a method that groups populations to maximise the variance between groups, used by Zhang Q. and colleagues.
Narrow-leaflet and broad-leaflet drug types: the groups of Lynch and colleagues (2016).
Basal: the name Ren and colleagues give to feral plants and landraces, mostly from China, closest to the ancestral gene pool.
Marijuana: the term in several of these sources for the samples called drug-type on the article.