Abstract
We propose a class of evolution models that involves an arbitrary exchangeable process as the breeding process and different selection schemes. In those models, a new genome is born according to the breeding process, and after that a genome is removed according to the selection scheme that involves fitness. Thus, the population size remains constant. The process evolves according to a Markov chain, and, unlike in many other existing models, the stationary distribution – so called mutation-selection equilibrium – can easily found and studied. As a special case our model contains a (sub) class of Moran models. The behaviour of the stationary distribution when the population size increases is our main object of interest. Several phase-transition theorems are proved.
Similar content being viewed by others
References
Baluja S (1997) Genetic algorithms and explicit search statistics. In: Advances in neural information processing systems, pp 319–325
Baluja S, Caruana R (1995) Removing the genetics from the standard genetic algorithm. In: ICML. Morgan Kauffman Publishers, Inc., pp 38–46
Baum EB, Boneh D, Garrett C (2001) Where genetic algorithms excel. Evol Comput 9(1):93–124
Bialowons R, Baake E (2008) Ancestral processes with selection: Branching and Moran models. Banach Center Publications, pp 80
Crow JF, Kimura M (1970) An introduction to population genetics theory. Harper & Row, Publishers, New York
Durrett R (2008) Probability models for DNA sequence evolution. Springer Science & Business Media
Elliott L, Teh YW (2012) Scalable imputation of genetic data with a discrete fragmentation-coagulation process. In: Advances in neural information processing systems, vol 25, pp 2861–2869
Etheridge AM, Griffiths RC (2009) A coalescent dual process in a Moran model with genic selection. Theor Popul Biol 75(4):320–330
Ewens WJ (2004) Mathematical population genetics. 2nd edn. Springer, New York
Feng S (2010) The Poisson-Dirichlet distribution and related topics: models and asymptotic behaviors. Springer Science & Business Media
Geman S, Geman D (1984) Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images. IEEE Trans Pattern Anal Mach Intell (6):721–741
Goldberg DE (1989) Genetic algorithms in search, optimization, and machine learning Addison-Wesley
Haario H, Saksman E (1991) Simulated annealing process in general state space. Adv Appl Probab 23(4):866–893
Hjort NL, Holmes C, Muller P, Walker S (2010) Bayesian nonparametrics. Number 28. Cambridge University Press
Holland JH (1975) Adaptation in natural and artificial systems. University of Michigan press
Kingman JFC (1980) Mathematics of genetic diversity, vol 34. SIAM
Kschischang FR, Frey BJ, Loeliger H -A (2001) Factor graphs and the sum-product algorithm. IEEE Trans Inf Theory 47(2):498–519
Moran PAP (1962) The statistical processes of evolutionary theory. Clarendon Press; Oxford University Press
Morse G, Stanley KO (2016) Simple evolutionary optimization can rival stochastic gradient descent in neural networks. In: proceedings of the genetic and evolutionary computation conference 2016. ACM, pp 477–484
Muirhead CA, Wakeley J (2009) Modeling multiallelic selection using a Moran model. Genetics 182(4):1141–1157
Neal RM (2000) Markov chain sampling methods for Dirichlet process mixture models. J Comput Graph Stat:249–265
Schrempf D, Hobolth A (2017) An alternative derivation of the stationary distribution of the multivariate neutral Wright–Fisher model for low mutation rates with a view to mutation rate estimation from site frequency data. Theor Popul Biol 114:88–94
Teh YW, Jordan MI (2010) Hierarchical Bayesian nonparametric models with applications. In: Bayesian nonparametrics. Cambridge University Press, pp 158–207
Vogl C, Clemente F (2012) The allele-frequency spectrum in a decoupled Moran model with mutation, drift, and directional selection, assuming small mutation rates. Theor Popul Biol 81(3):197–209
Vose MD (1999) The simple genetic algorithm: foundations and theory. The MIT Press
Watkins C, Buttkewitz Y (2014) Sex as Gibbs sampling: a probability model of evolution. arXiv:1402.2704
Acknowledgments
Research by Jüri Lember is supported by Estonian Institutional research funding IUT34-5 and PRG 865. Research by Chris Watkins is supported by grant number (FQXi Grant number FQXi-RFP-IPW-1913) from the Foundational Questions Institute and Fetzer Franklin Fund, a donor advised fund of Silicon Valley Community Foundation.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
Proof of claim 4.1
First note that
Now use the fact that if fn, f, gn, g are nonnegative functions such that \(\sup _{x} |f_{n}(x)-f(x)|\to 0\), \(\sup _{x} |g_{n}(x)-g(x)|\to 0\), \(\inf _{x} g(x)=g_{*}>0\) and f, g are bounded above, then with \(g^{*}=\sup _{x} g(x)\) and \(f^{*}=\sup _{x} f(x)\)
because for n big enough \(g_{n}(x)g(x)\geq {g^{2}_{*}\over 2}\) for every x. Take x = q, fn(q) = wn(k)q(k), f(q) = w(k)q(k), gn(q) = 〈wn, q〉 and g(q) = 〈w, q〉. Then g∗ = w(K) > 0, f∗ = g∗ = w(1) and so Eq. 4.8 follows. □
Proof of Proposition 5.1
Recall that fn and f are continuous and bounded functions on \(\mathcal {P}\) so that \(\|f_{n}\|_{\infty }<\infty \) and \(\|f\|_{\infty }<\infty \). By assumption, π is a finite measure. Since fn converges to f uniformly, it follows that \(\|f_{n}-f\|_{\infty }\to 0\) and so \(\| f_{n}\|_{\infty }\to \| f\|_{\infty }.\) For every m,
Since \(\|f\|_{m_{n}}\to \|f\|_{\infty }\), we have
Now fix δ > 0. Since \({\mathcal {P}^{*}_{\delta }}:=\{q: f(q) > \|f\|_{\infty }-\delta \}\), we have \({\mathcal {P}-\mathcal {P}^{*}_{\delta }}=\{q: f(q)\leq \|f\|_{\infty }-\delta \}. \) Define \(\delta ^{\prime }:=\delta /\|f\|_{\infty }\). Then
provided n is big enough. Thus,
so that \(\nu _{n}(\mathcal {P}^{*}_{\delta })\to 1\). We now argue that when \(\mathcal {P}^{*}=\{q^{*}\}\), then for any 𝜖 > 0 there exists δ > 0 so that
where B(q∗, 𝜖) is a ball in Euclidean sense. If, for an 𝜖 > 0, such a δ > 0 would not exists, then there would exist a sequence qn → q such that f(qn) ↗ f(q), but ∥qn − q∥≥ 𝜖. Since \(\mathcal {P}\) is compact, along a subsequence \(q_{n^{\prime }}\to q\) and by continuity \(f(q_{n^{\prime }})\to f(q)\). On the other hand ∥q − q∗∥ > 𝜖 and that would contradict the uniqueness of q∗. Therefore Eq. A.1 holds and so for any 𝜖 > 0, it holds that νn(B(q∗, 𝜖)) → 1. From the definition of the weak convergence, it now follows that \(\nu _{n}\Rightarrow \delta _{q^{*}}.\)□
Proof of Lemma 6.1
- 1):
-
To find
$$ q^{*}=\arg\max_{q\in \mathcal{P}} \left[ \ln \langle e^{-\phi} ,q \rangle + \sum\limits_{k}\alpha_{k}\ln q(k)\right], $$(A.2)we define Lagrangian
$$L(q,\beta)=\ln \langle e^{-\phi} ,q \rangle + \sum\limits_{k}\alpha_{k}\ln q(k)-\beta \left( \sum\limits_{k} q(k)-1\right)$$(here β is a scalar) and maximize L(q, β) over q > 0 (all entries are positive). Taking partial derivatives with respect to q(k), we have
$${e^{-\phi(k)}\over \langle e^{-\phi} ,q \rangle}+{\alpha_{k}\over q(k)}=\beta,\quad \Rightarrow \quad {e^{-\phi(k)}q(k)\over \langle e^{-\phi} ,q \rangle}+{\alpha_{k}}=q(k) \beta \quad \forall k.$$With \(|\alpha |={\sum }_{k} \alpha _{k},\) we have thus β = 1 + |α| and so the solution q∗ satisfies the set of equalities
$$ q^{*}(k)={1\over 1+|\alpha|}\left( {e^{-\phi(k)}q^{*}_{k}\over \langle e^{-\phi} ,q^{*} \rangle}+\alpha_{k}\right),\quad \forall k. $$(A.3)Now with w(k) = e−ϕ(k) define parameter θ := 〈w, q∗〉 and rewrite (A.3) as follows
$$ q^{*}(k)={\alpha_{k} \over (1+|\alpha|)-{w(k)\over \theta}} \quad k=1,\ldots,K. $$(A.4)We see that amongst the probability vectors satisfying 〈w, q∗〉 = θ, the solution is unique. Since αk > 0 for every k, it is easy to see that there is only one parameter θ such that the right hand side of Eq. A.4 would be a probability measure: if \(\theta ^{\prime }>\theta \), then for every k, we have
$${\alpha_{k} \over (1+|\alpha|)-{w(k)\over \theta}}>{\alpha_{k} \over (1+|\alpha|)-{w(k)\over \theta^{\prime}}}.$$Therefore a solution of Eq. A.2 is unique vector q∗ given by Eq. A.4, where θ = 〈w, q∗〉.
- 2):
-
To find
$$ q^{*}=\arg\max_{q\in \mathcal{P}} \left[-\langle \phi,q\rangle + \sum\limits_{k}\alpha_{k} \ln q(k)\right], $$(A.5)we define Lagrangian
$$L(q,\beta)= -\langle \phi,q\rangle + \sum\limits_{k}\alpha_{k}\ln q(k)-\beta \left( \sum\limits_{k} q(k)-1\right).$$Partial derivatives with respect to q(k) give us the equalities
$$ -\phi(k)+{\alpha_{k}\over q(k)}=\beta \quad \forall k\quad \Rightarrow \quad -\langle \phi,q\rangle + |\alpha| =\beta. $$Therefore, the inequalities for q∗(k) are
$$ q^{*}(k)={\phi(k)q^{*}(k) - \alpha_{k} \over \langle \phi,q^{*}\rangle -|\alpha|}={\phi(k)q^{*}(k) - \alpha_{k} \over \theta -|\alpha|},\quad \theta:=\langle \phi,q^{*}\rangle. $$(A.6)After rewriting (A.6), we obtain
$$ q^{*}(k)={ \alpha_{k} \over \phi(k)+|\alpha|-\theta},\quad k=1,\ldots,K, $$Thus, there cannot be two solutions having the same θ. As in the case 1), it is easy to see that when αk > 0 there is only one θ so that q∗ in Eq. A.2 sums up to one. Therefore, the solution to the problem (A.5) is unique. Note that the solution is independent of λ.
□
Rights and permissions
About this article
Cite this article
Lember, J., Watkins, C. An Evolutionary Model that Satisfies Detailed Balance. Methodol Comput Appl Probab 24, 1–37 (2022). https://doi.org/10.1007/s11009-020-09835-5
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11009-020-09835-5
Keywords
- Markov chain Monte Carlo
- Dirichlet distribution
- De Finetti theorem
- Weak convergence of probability measures
- Moran model