Data from: Enhanced computational methods for quantifying the effect of geographic and environmental isolation on genetic differentiation

In a recent paper, Bradburd et al. (2013) proposed a model to quantify the relative effect of geographic and environmental distance on genetic differentiation. Here, we enhance this method in several ways. 1. We modify the covariance model so as to fit better with mainstream geostatistical models and avoid mathematically ill-behaved covariance functions. 2. We extend the model-initially implemented only for co-dominant bi-allelic markers such as SNPs-to encompass highly polymorphic markers such as microsatellites 3. We implement and test a model selection procedure that allows users to assess which model (e.g. with or without an environment effect) is most suited. 4. We code all our MCMC algorithms in a mix of compiled languages which allows us to decrease computing time by at least one order of magnitude. 5. We propose an approximate inference and model selection method allowing us to deal with genomic datasets (several hundred thousands loci). 6. We also illustrate the potential of the method by re-analysing three datasets, namely harbour porpoises in Europe, coyotes in California and herrings in the Baltic Sea. The computer program developed here is freely available as an R package called Sunder. It takes as input geo-referenced allele counts at the individual or population level for co-dominant markers. Program homepage: www2.imm.dtu.dk/~gigu/Sunder.

近期，Bradburd等人（2013）在其发表的论文中提出了一种模型，用于量化地理距离与环境距离对遗传分化的相对影响。本研究从多个方面对该方法进行了改进：
1. 修正协方差模型，使其更贴合主流地统计模型，并规避病态协方差函数；
2. 扩展模型适用范围：该模型最初仅适用于单核苷酸多态性（SNPs）这类共显性双等位基因标记，现可覆盖微卫星等高度多态性标记；
3. 开发并测试模型选择流程，使研究者能够评估哪种模型（例如是否包含环境效应）最为适配；
4. 采用编译型语言混合编程实现所有马尔可夫链蒙特卡洛（Markov Chain Monte Carlo, MCMC）算法，将计算时间至少缩短一个数量级；
5. 提出近似推断与模型选择方法，可处理数十万位点的基因组数据集；
6. 通过重新分析三类数据集验证了该方法的应用潜力，分别为欧洲港湾鼠海豚、加州郊狼以及波罗的海鲱鱼相关数据集。

本研究开发的计算机程序以R语言包Sunder的形式免费发布，支持以共显性标记的个体或群体水平的地理参考等位基因计数作为输入。程序主页：www2.imm.dtu.dk/~gigu/Sunder。