概述

此页面提供了创建距离矩阵和运行和绘制非度量多维标度 (NMDS) 排序的基本代码。

在维基百科上了解更多关于排序的信息。

此代码依赖于 Jari Oksanen 在 R 中编写的 vegan 包。 Jari Oksanen.

数据

首先，导入数据并加载所需的库

require(MASS)
require(vegan)
data(varespec)   # species data
data(varechem)   # environmental data

距离矩阵

bray <- vegdist(varespec, method = "bray")				# calculate a distance matrix

# There are many distance measure options for 'dist', 
# discoverable by running '?dist'. Common distance measures include:
       # 'bray' = Bray-Curtis
       # 'canb' = Canberra
       # 'euclidean' = Euclidean

非约束排序

使用 NMDS 显示差异性

NMDS 分析和绘图

nmds <- metaMDS(varespec, k = 2, 
          distance = 'bray', autotransform = FALSE) 	# semi-black box NMDS function

ordiplot(nmds, type = "text")			      # Plot NMDS ordination
fit <- envfit(nmds, varechem[ ,1:4])			   # Calculates environmental vectors
fit						        # Lists vector endpoint coordinates and r-squared values
plot(fit)						   # adds environmental vectors
# a linear representation of environmental variables is not always appropiate
# we could also add a smooth surface of the variable to the plot
ordisurf(nmds, varechem$N, add = TRUE, col = "darkgreen")
nmds$stress                                             # stress value

在 metaMDS 函数中，k 是用户定义的，它与将投影限制在 k 维时投影拟合数据框的难易程度相关。传统观点认为应力不应超过 10-12%。应力可以通过增加维度来降低。但是，增加维度可能会降低二维图中前两个 NMDS 轴的“真实性”。

我们还可以运行一个具有 3 维的 nMDS，拟合环境向量并创建一个动态图形

nmds3d <- metaMDS(varespec, k = 3, 
  distance = 'bray', autotransform = FALSE)              # run nmds with 3 dimensions
nmds3d$stress                                            # stress drops
fit3d <- envfit(nmds3d, varechem[ ,1:4], choices = 1:3)  # fit environmental vectors to 3d space
ordirgl(nmds3d, envfit = fit3d)                          # dynamic 3D graph

对环境数据运行主成分分析 (PCA)

chem_pca <- rda(varechem, scale = TRUE)    # Run PCA
biplot(chem_pca, scaling = 2)              # display biplot

约束排序