通常将星星放在条形图或箱线图上以显示一组或两组之间的显着性水平(p值),以下是几个例子:
恒星的数量由p值定义,例如,可以将3个星形用于p值< 0.001,p值为2的两星。 0.01,依此类推(虽然这从一篇文章变为另一篇文章)。
我的问题:如何生成类似的图表?根据显着性水平自动放置星星的方法非常受欢迎。
答案 0 :(得分:35)
请在下面找到我的尝试。
首先,我创建了一些虚拟数据和一个可以根据需要修改的条形图。
windows(4,4)
dat <- data.frame(Group = c("S1", "S1", "S2", "S2"),
Sub = c("A", "B", "A", "B"),
Value = c(3,5,7,8))
## Define base plot
p <-
ggplot(dat, aes(Group, Value)) +
theme_bw() + theme(panel.grid = element_blank()) +
coord_cartesian(ylim = c(0, 15)) +
scale_fill_manual(values = c("grey80", "grey20")) +
geom_bar(aes(fill = Sub), stat="identity", position="dodge", width=.5)
如同baptiste已经提到的那样,在列上方添加星号很容易。只需使用坐标创建data.frame
。
label.df <- data.frame(Group = c("S1", "S2"),
Value = c(6, 9))
p + geom_text(data = label.df, label = "***")
要添加指示子组比较的弧,我计算了半圆的参数坐标,并将它们与geom_line
连接起来。星号也需要新的坐标。
label.df <- data.frame(Group = c(1,1,1, 2,2,2),
Value = c(6.5,6.8,7.1, 9.5,9.8,10.1))
# Define arc coordinates
r <- 0.15
t <- seq(0, 180, by = 1) * pi / 180
x <- r * cos(t)
y <- r*5 * sin(t)
arc.df <- data.frame(Group = x, Value = y)
p2 <-
p + geom_text(data = label.df, label = "*") +
geom_line(data = arc.df, aes(Group+1, Value+5.5), lty = 2) +
geom_line(data = arc.df, aes(Group+2, Value+8.5), lty = 2)
最后,为了表明群体之间的比较,我建立了一个更大的圆圈并将其展平在顶部。
r <- .5
x <- r * cos(t)
y <- r*4 * sin(t)
y[20:162] <- y[20] # Flattens the arc
arc.df <- data.frame(Group = x, Value = y)
p2 + geom_line(data = arc.df, aes(Group+1.5, Value+11), lty = 2) +
geom_text(x = 1.5, y = 12, label = "***")
答案 1 :(得分:34)
我知道这是一个老问题,Jens Tierling的答案已经为这个问题提供了一个解决方案。但我最近创建了一个ggplot扩展,简化了添加显着性条的整个过程:ggsignif
您只需添加单个图层geom_line
,而不是繁琐地将geom_text
和geom_signif
添加到您的图表中:
library(ggplot2)
library(ggsignif)
ggplot(iris, aes(x=Species, y=Sepal.Length)) +
geom_boxplot() +
geom_signif(comparisons = list(c("versicolor", "virginica")),
map_signif_level=TRUE)
要创建类似于Jens Tierling所示的更高级的绘图,您可以执行以下操作:
dat <- data.frame(Group = c("S1", "S1", "S2", "S2"),
Sub = c("A", "B", "A", "B"),
Value = c(3,5,7,8))
ggplot(dat, aes(Group, Value)) +
geom_bar(aes(fill = Sub), stat="identity", position="dodge", width=.5) +
geom_signif(stat="identity",
data=data.frame(x=c(0.875, 1.875), xend=c(1.125, 2.125),
y=c(5.8, 8.5), annotation=c("**", "NS")),
aes(x=x,xend=xend, y=y, yend=y, annotation=annotation)) +
geom_signif(comparisons=list(c("S1", "S2")), annotations="***",
y_position = 9.3, tip_length = 0, vjust=0.4) +
scale_fill_manual(values = c("grey80", "grey20"))
该软件包的完整文档可在CRAN获得。
答案 2 :(得分:17)
还有一个名为ggsignif的ggpubr包的扩展名,在多组比较方面更为强大。它建立在ggsignif之上,但也处理anova和kruskal-wallis以及与gobal均值的成对比较。
示例:
library(ggpubr)
my_comparisons = list( c("0.5", "1"), c("1", "2"), c("0.5", "2") )
ggboxplot(ToothGrowth, x = "dose", y = "len",
color = "dose", palette = "jco")+
stat_compare_means(comparisons = my_comparisons, label.y = c(29, 35, 40))+
stat_compare_means(label.y = 45)
答案 3 :(得分:2)
制作我自己的功能:
ts_test <- function(dataL,x,y,method="t.test",idCol=NULL,paired=F,label = "p.signif",p.adjust.method="none",alternative = c("two.sided", "less", "greater"),...) {
options(scipen = 999)
annoList <- list()
setDT(dataL)
if(paired) {
allSubs <- dataL[,.SD,.SDcols=idCol] %>% na.omit %>% unique
dataL <- dataL[,merge(.SD,allSubs,by=idCol,all=T),by=x] #idCol!!!
}
if(method =="t.test") {
dataA <- eval(parse(text=paste0(
"dataL[,.(",as.name(y),"=mean(get(y),na.rm=T),sd=sd(get(y),na.rm=T)),by=x] %>% setDF"
)))
res<-pairwise.t.test(x=dataL[[y]], g=dataL[[x]], p.adjust.method = p.adjust.method,
pool.sd = !paired, paired = paired,
alternative = alternative, ...)
}
if(method =="wilcox.test") {
dataA <- eval(parse(text=paste0(
"dataL[,.(",as.name(y),"=median(get(y),na.rm=T),sd=IQR(get(y),na.rm=T,type=6)),by=x] %>% setDF"
)))
res<-pairwise.wilcox.test(x=dataL[[y]], g=dataL[[x]], p.adjust.method = p.adjust.method,
paired = paired, ...)
}
#Output the groups
res$p.value %>% dimnames %>% {paste(.[[2]],.[[1]],sep="_")} %>% cat("Groups ",.)
#Make annotations ready
annoList[["label"]] <- res$p.value %>% diag %>% round(5)
if(!is.null(label)) {
if(label == "p.signif"){
annoList[["label"]] %<>% cut(.,breaks = c(-0.1, 0.0001, 0.001, 0.01, 0.05, 1),
labels = c("****", "***", "**", "*", "ns")) %>% as.character
}
}
annoList[["x"]] <- dataA[[x]] %>% {diff(.)/2 + .[-length(.)]}
annoList[["y"]] <- {dataA[[y]] + dataA[["sd"]]} %>% {pmax(lag(.), .)} %>% na.omit
#Make plot
coli="#0099ff";sizei=1.3
p <-ggplot(dataA, aes(x=get(x), y=get(y))) +
geom_errorbar(aes(ymin=len-sd, ymax=len+sd),width=.1,color=coli,size=sizei) +
geom_line(color=coli,size=sizei) + geom_point(color=coli,size=sizei) +
scale_color_brewer(palette="Paired") + theme_minimal() +
xlab(x) + ylab(y) + ggtitle("title","subtitle")
#Annotate significances
p <-p + annotate("text", x = annoList[["x"]], y = annoList[["y"]], label = annoList[["label"]])
return(p)
}
library(ggplot2);library(data.table);library(magrittr);
df_long <- rbind(ToothGrowth[,-2],data.frame(len=40:50,dose=3.0))
df_long$ID <- data.table::rowid(df_long$dose)
ts_test(dataL=df_long,x="dose",y="len",idCol="ID",method="wilcox.test",paired=T)
答案 4 :(得分:1)
我发现this one是有用的。
library(ggplot2)
library(ggpval)
data("PlantGrowth")
plt <- ggplot(PlantGrowth, aes(group, weight)) +
geom_boxplot()
add_pval(plt, pairs = list(c(1, 3)), test='wilcox.test')