如果没有时间精力学习代码，推荐了解：零代码数据挖掘课程

pROC package

以下是本包中常用的一些缩写

ROC: receiver operating characteristic,ROC曲线

AUC: area under the ROC curve,曲线下面积

pAUC: partial area under the ROC curve 部分曲线下面积

CI: confidence interval 可信区间

SP: specificity 特异度

SE: sensitivity 灵敏度

require(pROC)

data(aSAH)

if(!require(DT)) install.packages(DT)

DT::datatable(aSAH)

aSAH[1:5,1:5]

image.png

roc函数建立roc曲线

支持在管道中运行

参数分别为data, event, predict marker

library(dplyr)

aSAH %>%

filter(gender == "Female") %>%

roc(outcome, s100b)

Call:

roc.data.frame(data = ., response = outcome, predictor = s100b)

Data: s100b in 50 controls (outcome Good) < 21 cases (outcome Poor).

Area under the curve: 0.72

coords函数中筛选有效的的坐标

transpose参数指返回值的格式,FALSE 为row

这样筛选出了敏感度和特异度>0.6的坐标

library(dplyr)

aSAH %>%

filter(gender == "Female") %>%

roc(outcome, s100b) %>%

coords(transpose=FALSE) %>%

filter(sensitivity > 0.6,

specificity > 0.6)

threshold specificity sensitivity

1 0.155 0.68 0.6666667

2 0.165 0.74 0.6666667

3 0.175 0.76 0.6666667

4 0.185 0.78 0.6666667

5 0.215 0.80 0.6666667

6 0.245 0.82 0.6666667

7 0.255 0.82 0.6190476

建立roc 对象的方法

# Build a ROC object and compute the AUC

roc(aSAH$outcome, aSAH$s100b)

roc(outcome ~ s100b, aSAH)

建立光滑曲线

# Smooth ROC curve

roc(outcome ~ s100b, aSAH, smooth=TRUE)

Call:

roc.formula(formula = outcome ~ s100b, data = aSAH, smooth = TRUE)

Data: s100b in 72 controls (outcome Good) < 41 cases (outcome Poor).

Smoothing: binormal

Area under the curve: 0.74

可信区间与绘图

# more options, CI and plotting

roc1

aSAH$s100b, percent=TRUE,

# arguments for auc

partial.auc=c(100, 90), partial.auc.correct=TRUE,

partial.auc.focus="sens",

# arguments for ci

ci=TRUE, boot.n=100, ci.alpha=0.9, stratified=FALSE,

# arguments for plot

plot=TRUE, auc.polygon=TRUE, max.auc.polygon=TRUE, grid=TRUE,

print.auc=TRUE, show.thres=TRUE)

## 在原有图形上继续绘制

roc2

plot=TRUE, add=TRUE, percent=roc1$percent)

image.png

找出感兴趣的坐标

## Coordinates of the curve ##

coords(roc1, "best", ret=c("threshold", "specificity", "1-npv"),transpose = FALSE

)

coords(roc2, "local maximas", ret=c("threshold", "sens", "spec", "ppv", "npv"),transpose = FALSE)

threshold sensitivity specificity ppv npv

local.maximas -Inf 100.00000 0.00000 36.28319 NaN

local.maximas.1 1.5 95.12195 51.38889 52.70270 94.87179

local.maximas.2 2.5 65.85366 79.16667 64.28571 80.28169

local.maximas.3 3.5 63.41463 83.33333 68.42105 80.00000

local.maximas.4 4.5 43.90244 94.44444 81.81818 74.72527

local.maximas.5 Inf 0.00000 100.00000 NaN 63.71681

计算AUC可信区间

# CI of the AUC

ci(roc2)

95% CI: 74.85%-89.88% (DeLong)

plot在原有图形上增加

add=TRUE参数

roc1

aSAH$s100b, percent=TRUE,

# arguments for auc

partial.auc=c(100, 90), partial.auc.correct=TRUE,

partial.auc.focus="sens",

# arguments for ci

ci=TRUE, boot.n=100, ci.alpha=0.9, stratified=FALSE,

# arguments for plot

plot=TRUE, auc.polygon=TRUE, max.auc.polygon=TRUE, grid=TRUE,

print.auc=TRUE, show.thres=TRUE)

plot(roc2, add=TRUE)

image.png

比较AUC

看是否有统计学意义

# Test on the whole AUC

roc.test(roc1, roc2, reuse.auc=FALSE)

DeLong's test for two correlated ROC curves

data: roc1 and roc2

Z = -2.209, p-value = 0.02718

alternative hypothesis: true difference in AUC is not equal to 0

sample estimates:

AUC of roc1 AUC of roc2

73.13686 82.36789

绘制ROC曲线-基于ggplot2

创建roc对象

ggroc绘图

# Create a basic roc object

data(aSAH)

rocobj

rocobj2

绘图

基础绘图

library(ggplot2)

g

g

image.png

美化参数设置

ggroc(rocobj, alpha = 0.5, colour = "red", linetype = 2, size = 2)

image.png

支持gglot2语法的美化

# You can then your own theme, etc.

g + theme_minimal() + ggtitle("My ROC curve") +

geom_segment(aes(x = 1, xend = 0, y = 0, yend = 1), color="grey", linetype="dashed")

image.png

修改横纵坐标

# And change axis labels to FPR/FPR

gl

gl

gl + xlab("FPR") + ylab("TPR") +

geom_segment(aes(x = 0, xend = 1, y = 0, yend = 1), color="darkgrey", linetype="dashed")

image.png

image.png

绘制多条曲线

ggroc以list格式包裹roc对象

# Multiple curves:

g2

g2

image.png

也可先构建好公式，再绘制

# This is equivalent to using roc.formula:

roc.list

g.list

g.list

image.png

美化修改

size设置线条粗细

alpha设置透明度

# with additional aesthetics:

g3

g3+ggsci::scale_color_lancet()

image.png

改变参数

aes即按什么属性进行区分

g4

g4

image.png

按多种属性区分ROC曲线

# changing multiple aesthetics:

g5

g5

image.png

分面绘制ROC曲线

# OR faceting

g.list + facet_grid(.~name) + theme(legend.position="none")

image.png

所有曲线有相同颜色

group参数

# To have all the curves of the same color, use aes="group":

g.group

g.group

g.group + facet_grid(.~name)

image.png

我是白介素2，本期内容就到这里，下期再见。