TCGA泛癌TMB计算

在前面系列教程中分享了如何下载TCGA表达矩阵（2022新版TCGA数据下载与整理)，通过R进行数据合并就能得到TCGA泛癌表达矩阵了。有时候需要分析基因表达与TMB之间的相关性，就会涉及到TCGA突变数据下载和TMB计算。TCGA突变数据maf以前可以直接在TCGA下载，也可以用TCGAbiolinks包的GDCquery_Maf函数下载，后来TCGA不让下载了，TCGAbiolinks的GDCquery_Maf函数在新版本里面也被抛弃了。但是TCGAbiolinks包的作者提供了替代方案，具体见TCGAbiolinks: Searching, downloading and visualizing mutation files，作者提供了hg38和hg19两个版本的下载方法，以hg38的TCGA-CHOL为例：

library("TCGAbiolinks")
query <- GDCquery(
    project = "TCGA-CHOL", 
    data.category = "Simple Nucleotide Variation", 
    access = "open", 
    legacy = FALSE, 
    data.type = "Masked Somatic Mutation", 
    workflow.type = "Aliquot Ensemble Somatic Variant Merging and Masking"
)
GDCdownload(query)
maf <- GDCprepare(query)

得到的maf就可以用于后续TMB计算了，TMB的计算参考这个链接（TCGA 数据分析实战 —— TMB 与免疫浸润联合分析）：

get_TMB <- function(file) {
  # 需要用到的列
  use_cols <- c(
    "Hugo_Symbol", "Variant_Classification", "Tumor_Sample_Barcode", 
    "HGVSc", "t_depth", "t_alt_count"
  )
  # 删除这些突变类型
  mut_type <- c(
    "5'UTR", "3'UTR", "3'Flank", "5'Flank", "Intron", "IGR",
    "Silent", "RNA", "Splice_Region"
  )
  # 读取文件
  df <- read_csv(file, col_select = use_cols)
  data <- df %>% subset(!Variant_Classification %in% mut_type) %>%
    # 计算 VAF
    mutate(vaf = t_alt_count / t_depth) %>%
    group_by(Tumor_Sample_Barcode) %>%
    summarise(mut_num = n(), TMB = mut_num / 30, MaxVAF = max(vaf))
  return(data)
}

为了得到泛癌的TMB数据，我将代码写成循环，并稍微做了些调整。

rm(list=ls())
library(TCGAbiolinks)
library(dplyr)
library(stringr)
projects <- getGDCprojects()$project_id
projects <- projects[grepl('^TCGA', projects, perl=TRUE)]
projects <- projects[order(projects)]

pan_TMB=list()
for (project in projects){
  query <- GDCquery(
    project = project, 
    data.category = "Simple Nucleotide Variation", 
    access = "open", 
    legacy = FALSE, 
    data.type = "Masked Somatic Mutation", 
    workflow.type = "Aliquot Ensemble Somatic Variant Merging and Masking"
  )
  GDCdownload(query)
  maf <- GDCprepare(query)
  
  ## TMB calculation
  # Code Reference https://zhuanlan.zhihu.com/p/394609586
get_TMB <- function(file) {
  use_cols <- c("Hugo_Symbol", "Variant_Classification", "Tumor_Sample_Barcode", 
                  "HGVSc", "t_depth", "t_alt_count")
  # 删除这些突变类型
  mut_type <- c("5'UTR", "3'UTR", "3'Flank", "5'Flank", "Intron", "IGR",
                  "Silent", "RNA", "Splice_Region")
  # 读取文件
  df <- select(file, use_cols)
  data <- df %>% subset(!Variant_Classification %in% mut_type) %>%
  # 计算 VAF
    mutate(vaf = t_alt_count / t_depth) %>%
    group_by(Tumor_Sample_Barcode) %>%
    summarise(mut_num = n(), TMB = mut_num / 30, MaxVAF = max(vaf))
  return(data)
  }
  
  pan_TMB[[project]]=get_TMB(maf)
}

pan_TMB=do.call(rbind,pan_TMB)
pan_TMB$Cancer=str_split(rownames(pan_TMB),'-',simplify = T)[,2]
pan_TMB$Cancer=str_split(pan_TMB$Cancer,'[.]',simplify = T)[,1]
pan_TMB$Tumor_Sample_Barcode=str_sub(pan_TMB$Tumor_Sample_Barcode,1,16)
pan_TMB=pan_TMB[,c(5,1,3)]
colnames(pan_TMB)=c("Cancer","ID","TMB")
save(pan_TMB,file="pan_TMB.Rdata")

但是在循环到TCGA-MESO环节，总是会报"Error: Failure to retrieve index 140/0"，然后R就重启了。没办法，把TCGA-MESO单独出来，再合并。中间处理TCGA-MESO时在用maftools的read.maf时也有报错"Error in read.maf(x, isTCGA = TRUE) : No non-synonymous mutations found Check vc_nonSyn`` argumet inread.maffor details"，后来按照Github上的方法直接读入表格([No non-synonymous mutations found Check vc_nonSyn argumet in read.maf for details #838](https://github.com/PoisonAlien/maftools/issues/838)

最终代码如下：

rm(list=ls())
library(TCGAbiolinks)
library(dplyr)
library(stringr)
projects <- getGDCprojects()$project_id
projects <- projects[grepl('^TCGA', projects, perl=TRUE)]
projects <- projects[order(projects)]
projects=projects[-19]
pan_TMB=list()
for (project in projects){
  query <- GDCquery(
    project = project, 
    data.category = "Simple Nucleotide Variation", 
    access = "open", 
    legacy = FALSE, 
    data.type = "Masked Somatic Mutation", 
    workflow.type = "Aliquot Ensemble Somatic Variant Merging and Masking"
  )
  GDCdownload(query)
  maf <- GDCprepare(query)
  
  ## TMB calculation
  # Code Reference https://zhuanlan.zhihu.com/p/394609586
  get_TMB <- function(file) {
    use_cols <- c("Hugo_Symbol", "Variant_Classification", "Tumor_Sample_Barcode", 
                  "HGVSc", "t_depth", "t_alt_count")
    # 删除这些突变类型
    mut_type <- c("5'UTR", "3'UTR", "3'Flank", "5'Flank", "Intron", "IGR",
                  "Silent", "RNA", "Splice_Region")
    # 读取文件
    df <- select(file, use_cols)
    data <- df %>% subset(!Variant_Classification %in% mut_type) %>%
      # 计算 VAF
      mutate(vaf = t_alt_count / t_depth) %>%
      group_by(Tumor_Sample_Barcode) %>%
      summarise(mut_num = n(), TMB = mut_num / 30, MaxVAF = max(vaf))
    return(data)
  }
  
  pan_TMB[[project]]=get_TMB(maf)
}

# For TCGA-MESO
query <- GDCquery(
  project = "TCGA-MESO", 
  data.category = "Simple Nucleotide Variation", 
  access = "open", 
  legacy = FALSE, 
  data.type = "Masked Somatic Mutation", 
  workflow.type = "Aliquot Ensemble Somatic Variant Merging and Masking"
)
GDCdownload(query)

# Reference:https://github.com/PoisonAlien/maftools/issues/838
mafFilePath2 = dir(path = "GDCdata/TCGA-MESO",pattern = "masked.maf.gz$",full.names = T,recursive=T) 
TCGA_MESO = lapply(mafFilePath2, data.table::fread, skip = "Hugo_Symbol")
TCGA_MESO = data.table::rbindlist(l = TCGA_MESO, use.names = TRUE, fill = TRUE)
pan_TMB[["TCGA-MESO"]]=get_TMB(TCGA_MESO)

pan_TMB=do.call(rbind,pan_TMB)
pan_TMB$Cancer=str_split(rownames(pan_TMB),'-',simplify = T)[,2]
pan_TMB$Cancer=str_split(pan_TMB$Cancer,'[.]',simplify = T)[,1]
pan_TMB$Tumor_Sample_Barcode=str_sub(pan_TMB$Tumor_Sample_Barcode,1,16)
pan_TMB=pan_TMB[,c(5,1,3)]
colnames(pan_TMB)=c("Cancer","ID","TMB")
pan_TMB$TMB=round(pan_TMB$TMB,1) # 保留一位小数
pan_TMB=pan_TMB[order(pan_TMB$Cancer),]
save(pan_TMB,file="pan_TMB.Rdata")

最终数据如下：

image.png

查看下每个肿瘤的突变情况。

table(pan_TMB$Cancer)

# ACC BLCA BRCA CESC CHOL COAD DLBC ESCA  GBM HNSC KICH KIRC KIRP LAML  LGG LIHC LUAD LUSC MESO   OV PAAD PCPG 
# 90  414  986  289   51  454   47  184  461  510   66  357  282  128  521  371  616  544   78  462  170  182 
# PRAD READ SARC SKCM STAD TGCT THCA THYM UCEC  UCS  UVM 
 # 493  161  238  469  431  146  494  121  518   57   80 

通过ID可以和TCGA表达矩阵合并，再根据TCGA基因与免疫评分带侧边密度图的相关性点图的方法 就可以进行TMB与基因表达相关性分析了。