在R中生成矩阵中某些相关性的图形

Question

我想在变量(列)之间生成图形,这些图形具有高于和低于某个点的相关性以及p值<0.01.图表将是绘制相关的两列(变量)的ggplot2(线或条)图.

到目前为止,这是我的方法的要点,有一些虚拟数据,我会喜欢指向下一步的指针.

# Create some dummy data
df <- data.frame(sample(1:50), sample(1:50), sample(1:50), sample(1:50))
colnames(df) <- c("var1", "var2", "var3", "var4")

# Find correlations in the dummy data
df.cor <- cor(df)

# Make up some random pvalues for this example
x <- 0:1000
df.cor.pvals <- data.frame(sample(x/1000, 4), sample(x/1000, 4), sample(x/1000, 4), sample(x/1000,4))
colnames(df.cor.pvals) <- c("var1", "var2", "var3", "var4")

# Find the significant correlations
df.cor.extreme <- ((df.cor < -0.01 | df.cor > 0.01) & df.cor.pvals < 0.5)

# Ready data to for plotting
df$rownames <- rownames(df)
df.melt <- melt(df, id="rownames")

# I want to plot the combinations of variables that have a TRUE value
# in the df.cor.extreme matrix 

如果var1和var2的值为TRUE,则下面是硬编码示例.我假设这是我需要某种循环来生成多个图的地方,其中varA和varB是相关的.

ggplot(df.melt[(df.melt$variable=="var1" | df.melt$variable=="var2"),], aes(x=rownames, y=value, group=variable, colour=variable)) +
  geom_line()

Answer 1

如@DrewSteen的评论中所述,p-avlue必须与cor的形状相同.

在这里,我提供了一个计算p值矩阵的函数(它应该存在一个内置函数,在stats包中)

pvalue.matrix <- function(x,...){
  ncx <- ncol(x)
  r <- matrix(0, nrow = ncx, ncol = ncx)
  for (i in seq_len(ncx)) {
    for (j in seq_len(i)) {
      x2 <- x[, i]
      y2 <- x[, j]
      r[i, j] <-  cor.test(x2,y2,...)$p.value
    }
  }
  r <- r + t(r) - diag(diag(r))
  rownames(r) <- colnames(x)
  colnames(r) <- colnames(x)
  r
}

然后使用|的矢量化版本 并且像这样

df.cor.sig <- (df.cor > 0.01 | df.cor < -0.01) & pvalue.matrix(df) < 0.5

情节是经典的geom_tile

library(reshape2) ## melt
library(plyr)     ## round_any
 library(ggplot2) 
dat <- expand.grid(var1=1:4, var2=1:4)
dat$value <- melt(df.cor.sig)$value
dat$labels <- paste(round_any(df.cor,0.01) ,'(', round_any(pvalue.matrix(df),0.01),')',sep='')
ggplot(dat, aes(x=var1,y=var2,label=labels))+ 
  geom_tile(aes(fill = value),colour='white')+
 geom_text()

在OP澄清后编辑

plots <- apply(dat,1,function(x){
    plot.grob <- nullGrob()
    if(length(grep(pattern='TRUE',x[3])) >0 ){
      gg <- paste('var',c(x[1],x[2]),sep='')
      p <- ggplot(subset(df.melt,variable %in% gg ), 
            aes(x=rownames, y=value, group=variable, colour=variable)) +
            geom_line()
      plot.grob <- ggplotGrob(p)
    }
    plot.grob

})


library(gridExtra)
do.call(grid.arrange,  plots)