coalescents

Coalescents are a family of stochastic processes that model the genealogical relationships among a sample of gene lineages by tracing their ancestry backward in time. The most classical member is Kingman’s coalescent, introduced by John Kingman in 1982, which arises as a diffusion limit of the neutral Wright–Fisher model for large populations.

In Kingman’s model, when there are k lineages, pairs of lineages merge (coalesce) at rate binom(k,2) divided by 2N_e, where N_e is the effective population size. Time is typically rescaled in units of 2N_e generations, and the resulting genealogy is a binary tree with only pairwise mergers. The distribution of coalescence times determines the shape and depth of the ancestral tree, which in turn relates to patterns of genetic variation observed today.

Extensions of the basic coalescent relax the assumption of only pairwise mergers. Lambda-coalescents allow multiple lineages to merge in a single event, governed by a finite measure Lambda on [0,1]. Xi-coalescents further permit simultaneous multiple mergers with a random partition structure. These models are used to represent populations with highly skewed offspring distributions, such as certain marine species, pathogens, or populations with sweepstakes reproduction.

Applications and related concepts: When mutations occur along the genealogical tree, neutral models connect genealogies to observed DNA variation, enabling inference of demographic history and other population parameters. Coalescent theory underpins numerous simulation and inference tools (for example, msprime and related packages) and forms the basis for demographic skyline plots and Bayesian methods. Incorporating recombination leads to the ancestral recombination graph, which describes how genealogies vary along the genome; this adds substantial complexity to analysis and interpretation.

Limitations: The core Kingman coalescent assumes neutrality, constant population size, random mating, and no recombination or migration. Real data often violate these assumptions, prompting extensions that handle selection, structure, changing size, and recombination, though with greater methodological challenges.