如何对R数据框中各tax分组和指定列批量执行fisher.test检验

我们需要对给定数据框按以下规则执行批量统计检验：

• 分组维度：按tax字段的每个分组单独计算
• 检验列：对每个tax分组下的目标列（示例中为ABCB1、ABL1）分别执行fisher.test()检验
• 列联表构造规则：根据现有数据行计算生成，示例列联表格式如下：

计算逻辑：

• 42 = 总Present数 - tax1Present行对应ABCB1列的值（即43-1）
• 1 = tax1Present行对应ABCB1列的单元格值
• 20 = 总NotPresent数 - tax1NotPresent行对应ABCB1列的值（即23-3）
• 3 = tax1NotPresent行对应ABCB1列的单元格值

数据集构造代码：

df <- structure(list(group = c("tax1Present", "tax1NotPresent", "tax2Present", 
"tax2NotPresent", "tax3Present", "tax3NotPresent", "tax4Present", 
"tax4NotPresent", "tax5Present", "tax5NotPresent"), ABCB1 = c(1L, 
3L, 4L, 5L, 3L, 6L, 6L, 3, 2, 6L), ABL1 = c(18L, 14, 12L, 
9L, 1L, 5L, 0L, 0L, 7L, 0L), Present = c(43L, 43, 23L, 23, 
9L, 9, 7L, 7, 20, 20L),NotPresent = c(23, 23, 18, 18, 
7L, 7L, 10L, 10L, 10, 10L), tax = c("tax1", "tax1", "tax2", 
"tax2", "tax3", "tax3", "tax4", "tax4", "tax5", "tax5")), row.names = c(NA, 
10L), class = "data.frame")

数据集预览：

> df
            group ABCB1 ABL1 Present NotPresent  tax
1     tax1Present     1   18      43         23 tax1
2  tax1NotPresent     3   14      43         23 tax1
3     tax2Present     4   12      23         18 tax2
4  tax2NotPresent     5    9      23         18 tax2
5     tax3Present     3    1       9          7 tax3
6  tax3NotPresent     6    5       9          7 tax3
7     tax4Present     6    0       7         10 tax4
8  tax4NotPresent     3    0       7         10 tax4
9     tax5Present     2    7      20         10 tax5
10 tax5NotPresent     6    0      20         10 tax5

结合dplyr分组处理和purrr批量迭代即可完成计算，代码如下：

library(tidyverse)

# 指定要检验的目标列
target_cols <- c("ABCB1", "ABL1")

# 按tax分组批量计算
result <- df %>%
  group_by(tax) %>%
  group_modify(~{
    # 提取当前分组的两行数据
    present_row <- filter(.x, str_detect(group, "Present$"))
    not_present_row <- filter(.x, str_detect(group, "NotPresent$"))
    total_p <- present_row$Present
    total_np <- present_row$NotPresent
    
    # 对每个目标列计算fisher检验
    map_dfr(target_cols, function(col){
      # 构造列联表
      cont_table <- matrix(
        c(total_p - present_row[[col]], present_row[[col]],
          total_np - not_present_row[[col]], not_present_row[[col]]),
        nrow = 2, byrow = TRUE
      )
      # 执行fisher检验
      ft <- fisher.test(cont_table)
      # 返回结果
      tibble(
        target_col = col,
        p_value = ft$p.value,
        odds_ratio = as.numeric(ft$estimate),
        conf_low = ft$conf.int[1],
        conf_high = ft$conf.int[2]
      )
    })
  }) %>%
  ungroup()

# 查看结果
print(result)

运行后会输出每个tax分组、每个目标列对应的Fisher精确检验结果，包含p值、优势比及95%置信区间，可直接根据阈值筛选显著结果。

内容的提问来源于stack exchange，提问作者user2300940