Stochastic viability in an island model with partial dispersal: Approximation by a diffusion process in the limit of a large number of islands

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: We're looking at the paper: "Stochastic viability in an island model with partial dispersal".

Marcus: Here's the gist of the paper. The gist: As the number of demes approaches infinity, a discrete-time Markov chain modeling stochastic viability in an island model converges to a diffusion process for any deme size N ≥ 2, allowing for the evaluation of fixation probability.

Ines: Model Setup and Approximation The study considers a population structured into D demes, each containing N individuals of two types, A and B, where viability coefficients vary randomly from generation to generation The model uses a discrete-time Markov chain with two time scales to model the evolutionary process The authors assume that the means, variances and covariance of the viability coefficients are inversely proportional to the number of demes D, while higher-order moments are negligible in comparison to 1/D Diffusion Process Convergence The paper demonstrates that as the number of demes D approaches infinity, the accelerated Markov chain converges to a diffusion process for any deme size N ≥ 2 This diffusion process allows us to evaluate the fixation probability of type A following its introduction as a single mutant in a population that was fixed for type B The approximation relies on the existence of two timescales, a long one for the changes in strategy frequencies and a short one for deviations from equilibrium values Impact of Variability in Selection Coefficients The analysis explores the impact of increasing variability in the viability coefficients on fixation probability At least when N is large enough, it is shown that increasing this variability for type B or type A leads to an increase in the fixation probability of a single A The effect of the population-scaled variances, σ2A and σ2B, can even cancel the effects of the population-scaled means, µA and µB Moreover, this probability is higher for type A than for type B if the population-scaled geometric mean is higher for type A than for type B, which means that µA − σ2A/2 > µB − σ2B/2 Fixation Probability Results The fixation probability of type A introduced as a single mutant is approximated by FA ≈ 1/(ND) R1/0 S(y)dy, which simplifies to FA ≈ 1/(ND) R1/0 S(y)dy since Z(ND)-1/0 S(y)dy ≈ S(0)/ND = 1/ND Under the case where σ2A = σ2B = σAB, the fixation probability can be approximated as FA ≈ 2(µA − µB)/(ND)(1 − e−2(µA−µB)) if their population-scaled means are different Effect of Dispersal Rate The analysis shows that the fixation probability of A increases as the deme-scaled migration rate increases The effect of dispersal on FA is captured by M(x, y) and V(x, y), and increasing the deme-scaled dispersal rate will increase the fixation probability FA The paper concludes that selection favors the fixation of type A in a well-mixed population if µA > µB This conclusion is not valid with stochastic viability coefficients, as increasing the population-scaled variance of the viability coefficient of type B can increase the fixation probability of type A Furthermore, an increase in the deme-scaled dispersal rate will increase the fixation probability FA and FB Selection favors the type for which this geometric mean is the highest, i.e., µA − σ2A/2 > µB − σ2B/2 irrespective of the deme-scaled dispersal rate Final Remarks In the limit of a low deme-scaled dispersal rate, the effect of dispersal on the evolutionary process is negligible, and selection favors type A as long as µA − σ2A/2 > µB − σ2B/2 As ν tends towards infinity, the evolutionary process in the island model is analogous to the one in a well-mixed population.

Marcus: The effect of the population-scaled moments of the viability coefficients on the infinitesimal mean function is encapsulated in a difference of two terms where relatedness level is inversely proportional to the deme-scaled dispersal rate ν.

Ines: The effect of the population-scaled moments of the viability coefficients on the infinitesimal variance function is encapsulated in a difference between 1+σ2x(1−x) in the absence of relatedness and a correcting term for relatedness given by b y fIJσ2x(1−x).

Title and authors: Ines: The core of what this paper does is showing how to use this diffusion approximation to calculate that fixation probability of type A when it starts as a single mutant in a population fixed for type B. They use the convergence result we just talked about as the foundation for their analysis, which relies on having two distinct time scales to model the evolution.

Marcus: Right, so they are using that two-timescale approach to bridge the gap between their discrete-time Markov chain and this continuous diffusion process. This approximation is crucial because it lets them analyze how selection and dispersal interact in a population that isn't perfectly mixed.

Yuki: What this means for us in terms of real biology is that we can now look at scenarios where gene flow, or dispersal, plays a role, not just simple well-mixed dynamics. It connects the theoretical island model to the more realistic picture of populations partitioned into different groups subject to specific environmental pressures.

Ines: They also explore how increasing variability in those viability coefficients affects that fixation probability. They find that at least when N is large enough, increasing this variability for type B or type A actually leads to an increase in the fixation probability of a single A.

Marcus: That’s interesting because you’d expect selection to favor the fitter type, but here the randomness introduces a trade-off. They show that the effect of those population-scaled variances, sigma squared A and sigma squared B, can even cancel out the effects of those population-scaled means, mu A and mu B.

Yuki: And they find a specific condition where this cancellation doesn't happen—they show that the probability is higher for type A than for type B if the population-scaled geometric mean is higher for type A than for type B.

Ines: That condition is written as mu A minus sigma squared A over two greater than mu B minus sigma squared B over two, which simplifies things a bit when they set sigma squared A equals sigma squared B equals sigma AB. It’s a specific threshold based on the scaled means and variances.

Marcus: So, in short, they provide an analytical way to predict fixation probability under stochastic viability where structure and variability matter. But what's missing is how this maps onto the actual data we collect from these cohorts or island systems.

The paper's summary: Ines: Regarding the methodology itself, one major improvement they point out is their use of a discrete-time Markov chain with two time scales to model the evolutionary process. They emphasize that this structure is what allows them to infer weak convergence to a diffusion process in the limit of a large population, which they connect back to work by Ethier and Nagylaki on structured populations seven <ref:2311.00275#pg3,weak convergence to a diffusion process in the limit of a large>.

Marcus: And they're also leveraging the connection between models, showing that under certain conditions, like Fisher diffusion process, these different models become equivalent. This helps justify using the diffusion approximation over simpler models when you want to study migration and selection effects together.

Yuki: From a population perspective, their approach allows them to specifically focus on how geographic structure influences genetic diversity and its evolution in geographically structured natural populations, which is what we need when we look at species with distinct subpopulations.

Ines: They also show that the fixation probability of a single A increases as the deme-scaled migration rate increases. The effect of this dispersal on the fixation probability FA is captured by functions M(x, y) and V(x, y), and increasing that rate definitely boosts FA.

Marcus: That’s a concrete finding for us—dispersal isn't just background noise; it actively changes the outcome of selection in this island model. It shows that the ability to move between demes has a measurable impact on which type eventually dominates when viability is stochastic.

Yuki: So, if we think about real island populations or metapopulations, this tells us that dispersal isn't just about mixing; it’s a selective force that can help one strategy spread even if the local selection pressures are tricky.

The paper's improvements: Ines: To wrap up, the paper "Stochastic viability in an island model with partial dispersal: Approximation by a diffusion process in the limit of a large number of islands" shows us that selection favors type A as long as its population-scaled geometric mean is higher than that for type B, specifically when mu A minus sigma squared A over two is greater than mu B minus sigma squared B over two.

Marcus: And they also find that dispersal helps; increasing the deme-scaled migration rate increases the fixation probability of type A, so dispersal isn't just a passive feature here. It’s an active agent in the evolutionary dynamics described by this paper.

Yuki: I think what this really tells us is that in these structured settings, we have to look beyond simple well-mixed assumptions and account for how local structure interacts with selection variability, which is a key insight for understanding biodiversity across landscapes.

Ines: Exactly. So when we consider the full scope of this paper, it provides a diffusion process approximation to evaluate fixation probability in stochastic viability island models by showing that the geometric mean comparison dictates selection favorability regardless of dispersal rate in the limit of low dispersal.

Marcus: That means for our cohort studies, we have a theoretical framework to expect certain outcomes based on these scaled moments when we account for the structure and variability inherent in our sampling. It gives us something to test against what we actually observe in large datasets.

Yuki: It’s a solid piece of theory that connects population genetics history with the dynamics of structured environments, showing how local conditions dictate long-term evolutionary success under these stochastic rules.

Conclusion: Ines: So we just covered "Stochastic viability in an island model with partial dispersal: Approximation by a diffusion process in the limit of a large number of islands." We looked at how they use that two-timescale Markov chain to get to this continuous diffusion process and what that means for calculating fixation probability.

Marcus: Right, Ines, it’s about showing that as you have a big number of demes D approaching infinity, the discrete jump process smooths out into something we can actually calculate with diffusion math. It’s a way to handle the complexity of those two different time scales they set up for strategy changes versus equilibrium deviations.

Yuki: I wonder if this approximation holds up when you look at real island systems where dispersal isn't perfectly structured or where the number of demes is small, because that’s usually where the real biological signal comes from.

Ines: That’s a fair point, Yuki. The paper explicitly states this convergence happens for any deme size N greater than two, but it relies on those means and variances being inversely proportional to D, which is what lets them take that limit. It's a strong theoretical tool for the infinite limit.

Marcus: And what I found interesting about the results is how they show that increasing the variability in selection coefficients actually helps type A fixation if N is large enough, even when you have these population-scaled means and variances canceling each other out. That’s not just simple deterministic selection anymore.

Yuki: That cancellation thing is important because it shows that we can't always just look at the average viability; the noise in the environment itself can actually be a factor in which strategy wins out across different structural levels.

Ines: Exactly, Yuki. And they tie that back to this condition where mu A minus sigma squared A over two is greater than mu B minus sigma squared B over two. That geometric mean comparison becomes the deciding factor for selection favorability even when you factor in those complex stochastic effects from their viability coefficients.

Marcus: It’s a very specific condition for type A to win, and it doesn't even depend on the deme-scaled dispersal rate in that low dispersal limit they analyzed. That suggests selection strength, defined by those scaled moments, is the dominant force when migration is minimal.

Yuki: For a population geneticist, this means we can use these scaled moments to predict which strategy will spread across a landscape before we even have the full dataset for the whole system. It’s a powerful tool for making those predictions in complex spatial structures.

Ines: So, to wrap up, this paper shows how to use diffusion approximation on the discrete Markov chain from "Stochastic viability in an island model with partial dispersal" to get fixation probability estimates by focusing on the geometric mean comparison of scaled moments.

Marcus: It gives us a way to quantify how structural factors and selection variability play out over time in these models, even when you have migration terms involved.

Yuki: It’s a solid theoretical piece for understanding how spatial structure influences evolutionary outcomes in these types of models, and it opens up some avenues for testing those predictions against real island data.