Bridging Wright–Fisher and Moran models

Arthur Alexandre, Alia Abbara, Cecilia Fruet, Claude Loverdo, Anne‐Florence Bitbol

TL;DR

This paper introduces a new, flexible model of population genetics that bridges the gap between two classic models, the Wright-Fisher and Moran models. By allowing a tunable fraction of the population to be updated at each generation, the authors provide a unified framework to study allele frequency dynamics, fixation probabilities, and evolutionary timescales, offering a more nuanced view of population genetics than previously available.

What Problem Does This Paper Address?

Population genetics relies on mathematical models to understand how genetic variation changes over time within populations. Two cornerstone models are the Wright-Fisher (WF) model and the Moran model. The WF model assumes discrete, non-overlapping generations where an entire new generation is sampled from the previous one. The Moran model, in contrast, assumes overlapping generations with continuous birth-death events. While both models capture fundamental evolutionary forces like genetic drift and selection, they differ in their assumptions about generation structure, leading to distinct predictions, particularly regarding the efficiency of selection. This paper addresses the need for a more unified and flexible model that can interpolate between these two extremes, allowing researchers to explore scenarios that are not perfectly captured by either the WF or Moran model alone. This is particularly relevant for biological systems where generation overlap or update dynamics are intermediate.

Methods

The authors propose a novel model where a fixed fraction, denoted by $p$, of the population is updated at each discrete time step. This update involves randomly selecting $M$ individuals (where $M/N = p$) and replacing them with new individuals sampled from a binomial distribution. This framework naturally interpolates between the WF model (when $p=1$, the entire population is updated) and the Moran model (when $p=1/N$, effectively one individual is updated per step).

The core of their analysis relies on the diffusion approximation, a standard technique in population genetics that treats the change in allele frequency as a continuous process. They derive expressions for:

  • Fixation probability: The likelihood that a new advantageous mutation will become fixed in the population.
  • Absorption times: The average time it takes for a mutation to either fix or go extinct.
  • Coalescent process: The genealogy of sampled individuals tracing back in time, which converges to Kingman's coalescent under appropriate scaling.
  • Effective population sizes: Various measures quantifying how a population behaves genetically like a simpler WF population.

The authors also generalize their model to include fluctuating update fractions, fluctuating individual lifetimes, and selection acting on both reproductive fitness and lifetime.

Datasets

Not discussed in the paper. The study is theoretical and relies on mathematical derivations and simulations.

Key Insights and Ideas

  • A Unified Framework: The proposed model elegantly bridges the WF and Moran models by introducing a single parameter, the update fraction $p$. This allows for a continuous spectrum of generational overlap and update dynamics, offering a more realistic representation of many biological systems.
  • Intermediate Selection Efficiency: The fixation probability of beneficial mutations in the new model falls between the WF and Moran models, suggesting that selection is more efficient in the WF model than in the Moran model, with intermediate $p$ values yielding intermediate efficiency.
  • Coalescent Convergence: The model's coalescent process, under the correct time scaling, converges to Kingman's coalescent, a fundamental result in population genetics, reinforcing its generality.
  • Coinciding Effective Population Sizes: For this model, three different definitions of effective population size (variance, eigenvalue, and coalescent) coincide, simplifying theoretical analysis and providing a robust measure of the population's genetic behavior.
  • Impact of $p$ on Dynamics: The update fraction $p$ significantly influences fixation probabilities, absorption times, and effective population sizes, demonstrating its importance in shaping evolutionary trajectories.

Key Results

  • The fixation probability $\phi(x)$ for a mutant starting at frequency $x$ is given by a simple expression that depends on $p$. For a single mutant ($x=1/N$) and weak selection ($s \ll 1$), $\phi_1 = 2s/(2-\rho)$, where $\rho = p$. This interpolates between the Moran result ($\phi_1 = s$) and the WF result ($\phi_1 = 2s$).
  • Mean absorption times (fixation and extinction) are derived and shown to be dependent on $p$.
  • The rate of approach to steady state in the neutral case is characterized by an eigenvalue $\lambda^* = 1 - \rho(2-\rho)/N$, which interpolates between the Moran ($\rho=1/N$) and WF ($\rho=1$) limits.
  • The effective population size $N_e$ is found to be $N/(2-\rho)$, which increases non-linearly with $p$, ranging from $N/2$ (Moran) to $N$ (WF).
  • Generalizations to fluctuating update fractions and lifetimes yield similar analytical tractability, with the mean individual lifetime being the inverse of the mean update fraction.

Limitations and Future Work

  • No Mutation: The current model focuses on fixation and extinction and does not explicitly incorporate mutation. The authors suggest that including mutations would allow for the derivation of stationary distributions and would be an interesting extension.
  • Well-Mixed Population: The model assumes a well-mixed population. Extending it to spatially structured populations or considering fluctuations in population size are suggested future directions.
  • Diploid Case: The current work focuses on haploid populations. Extending the model to diploids would be a valuable next step.
  • Continuous Time: The model uses discrete time steps. While relevant for some biological systems, adapting it to continuous time would broaden its applicability.
  • Specific Distributions: While generalizations are explored for fluctuating lifetimes and update fractions, the specific distributions used (e.g., geometric for lifetimes) could be varied.

Summary updated

Abstract

The Wright-Fisher model and the Moran model are both widely used in population genetics. They describe the time evolution of the frequency of an allele in a well-mixed population with fixed size. We propose a simple and tractable model which bridges the Wright-Fisher and the Moran descriptions. We assume that a fixed fraction of the population is updated at each discrete time step. In this model, we determine the fixation probability of a mutant and its average fixation and extinction times, under the diffusion approximation. We further study the associated coalescent process, which converges to Kingman's coalescent, and we calculate effective population sizes. We generalize our model, first by taking into account fluctuating updated fractions or individual lifetimes, and then by incorporating selection on the lifetime as well as on the reproductive fitness.
BibTeX
@article{alexandre2024bridging,
  author = {Arthur Alexandre and Alia Abbara and Cecilia Fruet and Claude Loverdo and Anne‐Florence Bitbol},
  title = {{Bridging Wright–Fisher and Moran models}},
  journal = {Journal of Theoretical Biology},
  year = {2024},
  doi = {10.1016/j.jtbi.2024.112030},
}
Figure