[论文解读] Amplifiers and Suppressors of Selection for the Moran Process on Undirected Graphs

George Giakkoupis|arXiv (Cornell University)|Nov 5, 2016

Game Theory and Applications参考文献 31被引用 15

一句话总结

本文首次展示了已知的无向图在莫兰过程下可作为强放大器或强抑制器，证明随着图规模增大，固定概率可趋近于1或0。该构造简洁且可证明为最优，放大器的固定概率为 $1 - \tilde{O}(n^{-1/3})$，抑制器为 $\tilde{O}(n^{-1/4})$，推翻了长期以来认为无向图中此类图不存在的信念。

ABSTRACT

We consider the classic Moran process modeling the spread of genetic mutations, as extended to structured populations by Lieberman et al.\ (Nature, 2005). In this process, individuals are the vertices of a connected graph $G$. Initially, there is a single mutant vertex, chosen uniformly at random. In each step, a random vertex is selected for reproduction with a probability proportional to its fitness: mutants have fitness $r>1$, while non-mutants have fitness 1. The vertex chosen to reproduce places a copy of itself to a uniformly random neighbor in $G$, replacing the individual that was there. The process ends when the mutation either reaches fixation (i.e., all vertices are mutants), or gets extinct. The principal quantity of interest is the probability with which each of the two outcomes occurs. A problem that has received significant attention recently concerns the existence of families of graphs, called strong amplifiers of selection, for which the fixation probability tends to 1 as the order $n$ of the graph increases, and the existence of strong suppressors of selection, for which this probability tends to 0. For the case of directed graphs, it is known that both strong amplifiers and suppressors exist. For the case of undirected graphs, however, the problem has remained open, and the general belief has been that neither strong amplifiers nor suppressors exist. In this paper we disprove this belief, by providing the first examples of such graphs. The strong amplifier we present has fixation probability $1- ilde O(n^{-1/3})$, and the strong suppressor has fixation probability $ ilde O(n^{-1/4})$. Both graph constructions are surprisingly simple. We also prove a general upper bound of $1- ilde Ω(n^{-1/3})$ on the fixation probability of any undirected graph. Hence, our strong amplifier is existentially optimal.

研究动机与目标

• 解决无向图中是否存在强放大器或抑制器这一开放问题，挑战既有的‘此类图不存在’的普遍信念。
• 构建显式、简洁的无向图族，使其在莫兰过程中表现出极端的固定概率。
• 为无向图建立固定的理论紧界，证明所提出的放大器在存在性上的最优性。
• 探讨这些结构对演化动力学的影响，包括在体细胞演化与社会影响传播中的应用。

提出的方法

• 设计一种新颖的无向图构造方法，通过分层顶点分组与受控连通性，将固定偏向突变体。
• 通过修改连通性以隔离并限制突变体传播，构建互补的抑制器图。
• 使用概率分析与 hitting 时间的界来估计固定概率，依赖于顶点权重与邻域结构的调和平均。
• 应用集中不等式与尾部估计，控制图层间顶点贡献总和中的误差项。
• 利用对称性与均匀随机初始化，推导单突变体起始下的整体固定概率 $\rho_G(r)$。
• 证明任意无向图的固定概率上界为 $1 - \tilde{\Omega}(n^{-1/3})$，表明该放大器在存在性上是最优的。

实验结果

研究问题

• RQ1 在无向图的莫兰过程中，是否存在强放大器？
• RQ2 尽管存在相反的先验信念，是否可在无向图中构造出强抑制器？
• RQ3 无向图的固定概率理论极限是什么？能否实现该极限？
• RQ4 图的结构与连通性如何影响结构化种群中的固定动力学？
• RQ5 是否可构造出简单且可解析的图族，实现接近最优的固定行为？

主要发现

• 首次提出无向图中强放大器的显式构造，其固定概率为 $1 - \tilde{O}(n^{-1/3})$。
• 构造出固定概率为 $\tilde{O}(n^{-1/4})$ 的强抑制器，证明无向图中此类图确实存在。
• 所提出的放大器实现了渐近最优的固定概率，与理论上的上界 $1 - \tilde{\Omega}(n^{-1/3})$ 一致。
• 分析证实，无向图中既可实现选择的放大，也可实现抑制，推翻了先前的假设。
• 构造极为简洁，暗示其在真实演化或社交网络中具有潜在应用价值。
• 结果表明，可通过设计网络拓扑，实现对结构化种群中优势性状固定的促进或抑制。