Wright-Fisher Model

Wright-Fisher model is the canonical mathematical model of population genetics, developed independently by Sewall Wright and R.A. Fisher in the early 1930s. It describes how allele frequencies change from one generation to the next in an idealized population of constant size N, where each individual in the offspring generation is formed by randomly sampling two parents (with replacement) from the previous generation.

The model's simplifying assumptions — discrete generations, random mating, no selection, no mutation, no gene flow — make it analytically tractable but biologically unrealistic. Its value lies not in descriptive accuracy but in providing a null model against which real population dynamics can be measured. Deviations from Wright-Fisher predictions reveal the action of selection, drift, or non-random mating.

Under the Wright-Fisher model, the probability that an allele reaches fixation is equal to its initial frequency, and the expected time to fixation or loss scales with population size. The model is the foundation of coalescent theory and the diffusion approximation for allele frequency dynamics, which extend it to include selection and mutation.

The Wright-Fisher Model as a Null Hypothesis

The deepest systems-theoretic insight of the Wright-Fisher framework is not demographic but epistemological: it defines what randomness looks like in a finite population. The model does not claim that real populations are random-mating panmictic units of constant size. It claims that if we observe allele frequency changes that deviate from Wright-Fisher predictions, those deviations are evidence of structure — selection, population structure, non-random mating, or historical demography — that the null model does not capture.

This is the logic of null models in complex systems generally. A null model strips away structure to reveal what remains when structure is absent. The Erdős-Rényi random graph does not claim that social networks are random; it claims that deviations from randomness reveal the organizational principles of real networks. The ideal gas does not claim that gases have no intermolecular forces; it claims that deviations from ideality reveal the strength and range of those forces. The Wright-Fisher model operates in the same epistemological register: it is a deliberately wrong model whose wrongness is informative.

Mathematical Structure

The Wright-Fisher process is a Markov chain on the state space of allele frequencies. If there are N diploid individuals, there are 2N gene copies, and the state space is the set of integers {0, 1, 2, ..., 2N} representing the count of one allele. The transition probabilities follow a binomial distribution: the probability of k copies in generation t+1, given i copies in generation t, is:

P(X_{t+1} = k | X_t = i) = C(2N, k) * (i/2N)^k * (1 - i/2N)^{2N-k}

This is the genetic drift process: random sampling of parents with replacement. The absorbing states are 0 and 2N — fixation and loss — and the chain is guaranteed to eventually reach one of them. The expected time to absorption scales as O(N) for neutral alleles, and the probability of eventual fixation of a new mutation is 1/(2N).

The key insight is that genetic drift is not a force like selection. It is the absence of force — the signature of stochasticity in a finite system. The Wright-Fisher model makes this absence mathematically precise.

Extensions and Violations

Real populations violate Wright-Fisher assumptions in ways that are themselves informative:

• Population structure (the Wahlund effect): Subdivided populations show excess homozygosity because the global allele frequency is not the same as the local frequency. The deviation from Hardy-Weinberg equilibrium measures the degree of structure.

• Selection: The diffusion approximation extends Wright-Fisher to include selection by adding a directional term to the drift process. The relative magnitude of selection (s) and drift (1/2N) determines the evolutionary regime: when 4Ns >> 1, selection dominates; when 4Ns << 1, drift dominates.

• Overlapping generations: The Moran model replaces discrete generations with a continuous-time death-birth process. The Moran model and Wright-Fisher model agree in the diffusion limit but differ in finite-population behavior.

• Fluctuating population size: Real populations expand and contract. The effective population size Ne is the size of an ideal Wright-Fisher population that would exhibit the same amount of genetic drift as the real population. Ne is typically much smaller than the census size because bottlenecks, skewed reproductive success, and population structure all amplify drift.

The Wright-Fisher Model and the Philosophy of Randomness

The model raises a subtle question about the nature of randomness in biological systems. Is genetic drift really random, or is it merely unpredictable given our incomplete knowledge of the mating system, the environment, and the fitness landscape? The Wright-Fisher model sidesteps this question by operationalizing randomness: an allele frequency change is random if it is statistically indistinguishable from binomial sampling.

This operational definition has consequences. It implies that drift and selection are not mutually exclusive forces but limiting cases of a continuum. Every allele frequency change is driven by both deterministic (selection) and stochastic (drift) components, and the Wright-Fisher framework provides the null distribution against which the deterministic component can be detected.

The model also connects to broader questions in statistical mechanics. The ergodic hypothesis in physics assumes that time averages equal ensemble averages. In population genetics, the Wright-Fisher process is not ergodic: a single population trajectory will eventually fix or lose every allele, while the ensemble of populations maintains polymorphism. The timescale on which ergodicity breaks down — the absorption time — is the timescale on which evolution loses its memory of initial conditions.

The Wright-Fisher model is not a description of evolution. It is a description of what evolution would look like if nothing structured it. Its power lies in its poverty: by assuming almost nothing, it reveals what must be explained.