Data from: Bayesian inference of selection in a heterogeneous environment from genetic time-series data

Evolutionary geneticists have sought to characterize the causes and molecular targets of selection in natural populations for many years. Although this research program has been somewhat successful, most statistical methods employed were designed to detect consistent, weak to moderate selection. In contrast, phenotypic studies in nature show that selection varies in time and that individual bouts of selection can be strong. Measurements of the genomic consequences of such fluctuating selection could help test and refine hypotheses concerning the causes of ecological specialization and the maintenance of genetic variation in populations. Herein, I proposed a Bayesian non-homogenous hidden Markov model to estimate effective population sizes and quantify variable selection in heterogeneous environments from genetic time-series data. The model is described and then evaluated using a series of simulated data, including cases where selection occurs on a trait with a simple or polygenic molecular basis. The proposed method accurately distinguished neutral loci from non-neutral loci under strong selection, but not from those under weak selection. Selection coefficients were accurately estimated when selection was constant or when the fitness values of genotypes varied linearly with the environment, but these estimates were less accurate when fitness was polygenic or the relationship between the environment and the fitness of genotypes was non-linear. Past studies of temporal evolutionary dynamics in lab populations have been remarkably successful. The proposed method makes similar analyses of genetic time-series data from natural populations more feasible, and thereby could help answer fun damental questions about the causes and consequences of evolution in the wild.进化遗传学家多年来一直致力于解析自然种群中自然选择的成因与分子靶点。尽管该研究方向已取得一定进展，但目前主流的统计方法均仅能用于检测持续存在的弱至中等强度选择。与之形成鲜明对比的是，野外表型研究显示，自然选择的强度随时间波动，且单次选择事件可具有极强的作用强度。针对此类波动选择所引发的基因组层面效应开展定量测量，将有助于检验并完善关于生态特化的成因以及种群内遗传变异维持机制的相关假说。 本文提出一种贝叶斯非齐次隐马尔可夫模型（Bayesian non-homogenous hidden Markov model），用于从遗传时间序列数据（genetic time-series data）中估算有效种群大小（effective population size），并量化异质环境中的可变选择强度。本文先对该模型进行形式化描述，随后通过一系列模拟数据对其性能开展评估，其中涵盖选择作用于具有单基因或多基因分子基础的性状的场景。所提出的方法可准确区分受强选择的非中性位点（non-neutral loci）与中性位点（neutral loci），但无法区分受弱选择的非中性位点与中性位点。当选择作用恒定，或基因型（genotypes）适合度值（fitness values）随环境呈线性变化时，选择系数（selection coefficients）可被较为精准地估算；但当适合度由多基因调控，或环境与基因型适合度之间的关联呈非线性时，估算结果的准确性会有所下降。 过往针对实验室种群时间尺度进化动态的研究已取得显著成果。本文提出的方法可使针对自然种群遗传时间序列数据的类似分析更具可行性，从而助力解答关于野外进化的成因与效应的基础性科学问题。