GWAS數(shù)據(jù)沒有提供eaf，如何是好……

昵稱69125444 2023-09-12

展開全文

事情是這樣的，好不容易從文獻中找到一手SNPs數(shù)據(jù)，喜大普奔……

定睛一看，沒有eaf值啊，這可咋整，后續(xù)需要用到read_outcome_data函數(shù)，eaf值是必須的呢！

在這里停滯了好久，準(zhǔn)備放棄這部分數(shù)據(jù)了，但是又覺得很可惜，數(shù)次徘徊……

于是開始搜搜搜，然后B站還真的給我推了，柳暗花明又一村啊——

1方法1：snp_add_eaf

[孟德爾隨機化之代碼生成eaf_嗶哩嗶哩_bilibili https://www.bilibili.com/video/BV1kR4y1f7Pz/?spm_id_from=333.788.recommend_more_video.-1&vd_source=e5e36ce10569a7a5d647d18bdb42e4b5

更驚喜的事情是，原作者閃現(xiàn)評論區(qū)提供了代碼出處：

[(https://github.com/linjf15/MR_tricks/tree/main/GWAS_preprocessing)

snp_add_eaf <- function(dat, build = '37', pop = 'EUR')
{
  stopifnot(build %in% c('37','38'))
  stopifnot('SNP' %in% names(dat))
  
  # Create and get a url
  server <- ifelse(build == '37','http://grch37.rest.','http://rest.')
  pop <- paste0('1000GENOMES:phase_3:',pop)
  
  snp_reverse_base <- function(x)
  {
    x <- stringr::str_to_upper(x)
    stopifnot(x %in% c('A','T','C','G'))
    switch(x,'A'='T','T'='A','C'='G','G'='C')
  }
  
  res_tab <- lapply(1:nrow(dat), function(i)
  {
    print(paste0('seaching for No.', i, ' SNP'))
    dat_i <- dat[i,]
    
    ext <- paste0('/variation/Homo_sapiens/',dat_i$SNP, '?content-type=application/json;pops=1')
    url <- paste(server, ext, sep = '')
    res <- httr::GET(url)
    
    # Converts http errors to R errors or warnings
    httr::stop_for_status(res)
    
    # Convert R objects from JSON
    res <- httr::content(res)
    res_pop <- jsonlite::fromJSON(jsonlite::toJSON(res))$populations
    
    # Filter query results based on population set
    res_pop <- try(res_pop[res_pop$population == pop,])
    if('try-error' %in% class(res_pop))
    {
      print(paste0('There is not information for population ',pop))
      queried_effect_allele <- 'NR'
      queried_other_allele <- 'NR'
      queried_eaf <- -1
    }
    else
    {
      if(nrow(res_pop)==0)
      {
        print(paste0('There is not information for population ',pop))
        queried_effect_allele <- 'NR'
        queried_other_allele <- 'NR'
        queried_eaf <- -1
      }
      else
      {
        queried_effect_allele <- res_pop[1,'allele'][[1]]
        queried_other_allele <- res_pop[2,'allele'][[1]]
        queried_eaf <- res_pop[1,'frequency'][[1]]    
      }
    }
    
    effect_allele <- ifelse('effect_allele.exposure' %in% names(dat),
                            dat_i$effect_allele.exposure,
                            dat_i$effect_allele)
    
    other_allele <- ifelse('effect_allele.exposure' %in% names(dat),
                            dat_i$other_allele.exposure,
                            dat_i$other_allele)
    
    if('effect_allele.exposure' %in% names(dat))
    {
      name_output <- unique(c(names(dat), 'eaf.exposure','reliability.exposure'))
    }
    else
    {
      name_output <- unique(c(names(dat), 'eaf','reliability.exposure'))
    }
      
    len_effect_allele <- nchar(effect_allele)
    len_other_allele <- nchar(other_allele)
    
    if(len_effect_allele==1&len_other_allele==1)
    {
      if((queried_effect_allele==effect_allele & queried_other_allele==other_allele)|
         (queried_effect_allele==other_allele & queried_other_allele==effect_allele))
      {
        dat_i$eaf.exposure <- ifelse(effect_allele == queried_effect_allele,
                                     queried_eaf,
                                     1-queried_eaf)
        dat_i$eaf <- dat_i$eaf.exposure 
        dat_i$reliability.exposure <- 'high'
      }
      else
      {
        r_queried_effect_allele <- snp_reverse_base(queried_effect_allele)
        r_queried_other_allele <- snp_reverse_base(queried_other_allele)
        if((r_queried_effect_allele==effect_allele & r_queried_other_allele==other_allele)|
           (r_queried_effect_allele==other_allele & r_queried_other_allele==effect_allele))
        {
          dat_i$eaf.exposure <- ifelse(effect_allele == r_queried_effect_allele,
                                       queried_eaf,
                                       1-queried_eaf)
          dat_i$eaf <- dat_i$eaf.exposure 
          dat_i$reliability.exposure <- 'high'
        }
        else
        {
          dat_i$eaf.exposure <- ifelse(effect_allele == queried_effect_allele,
                                       queried_eaf,
                                       1-queried_eaf)
          dat_i$eaf <- dat_i$eaf.exposure 
          dat_i$reliability.exposure <- 'low'
        }
      }
    }
    
    else
    {
      # To identify the potential DEL/ INS
      short_allele <- ifelse(len_effect_allele==1,
                             effect_allele,
                             other_allele)
      short_allele_eaf <- ifelse(short_allele == queried_effect_allele, 
                                 queried_eaf, 
                                 1-queried_eaf)
      dat_i$eaf.exposure <- ifelse(effect_allele == short_allele,
                                   short_allele_eaf,
                                   1-short_allele_eaf)
      dat_i$eaf <- dat_i$eaf.exposure 
      dat_i$reliability.exposure <- 'low'
    }
    
    dat_i[name_output]
  })
  
  return(do.call(rbind, res_tab))
}

如果能確定數(shù)據(jù)的參考基因組版本，還可以自定義哦~

2方法2：get_eaf_from_1000G

[孟德爾隨機化, 補eaf, 本地與在線_嗶哩嗶哩_bilibili]
[(https://www.bilibili.com/video/BV1Dh411j7yP/?spm_id_from=333.999.0.0&vd_source=e5e36ce10569a7a5d647d18bdb42e4b5)

這個up主介紹了兩種方法，第一種如上所述，第二種來自于另外一個團隊【代碼作者：廣州醫(yī)科大學(xué)第一臨床學(xué)院周浩彬，第二臨床學(xué)院謝治鑫】，用起來都非常方便。

說明文檔：[Get_MR/Get_MR1.0 help.md at main · HaobinZhou/Get_MR · GitHub]
[(https://github.com/HaobinZhou/Get_MR/blob/main/Get_MR1.0 help.md)

get_eaf_from_1000G<-function(dat,path,type='exposure'){
  
  corrected_eaf_expo<-function(data_MAF){
    effect=data_MAF$effect_allele.exposure
    other=data_MAF$other_allele.exposure
    A1=data_MAF$A1
    A2=data_MAF$A2
    MAF_num=data_MAF$MAF
    EAF_num=1-MAF_num
    
    harna<-is.na(data_MAF$A1)
    harna<-data_MAF$SNP[which(harna==T)]
    
    cor1<-which(data_MAF$effect_allele.exposure !=data_MAF$A1)
    
    
    data_MAF$eaf.exposure=data_MAF$MAF
    data_MAF$type='raw'
    data_MAF$eaf.exposure[cor1]=EAF_num[cor1]
    data_MAF$type[cor1]='corrected'
    cor2<-which(data_MAF$other_allele.exposure ==data_MAF$A1)
    cor21<-setdiff(cor2,cor1)
    cor12<-setdiff(cor1,cor2)
    error<-c(cor12,cor21)
    data_MAF$eaf.exposure[error]=NA
    data_MAF$type[error]='error'
    
    data_MAF<-list(data_MAF=data_MAF,cor1=cor1,harna=harna,error=error)
    
    return(data_MAF)
    
  }
  
  corrected_eaf_out<-function(data_MAF){
    effect=data_MAF$effect_allele.outcome
    other=data_MAF$other_allele.outcome
    A1=data_MAF$A1
    A2=data_MAF$A2
    MAF_num=data_MAF$MAF
    EAF_num=1-MAF_num
    
    harna<-is.na(data_MAF$A1)
    harna<-data_MAF$SNP[which(harna==T)]
    
    cor1<-which(data_MAF$effect_allele.outcome !=data_MAF$A1)
    
    
    data_MAF$eaf.outcome=data_MAF$MAF
    data_MAF$type='raw'
    data_MAF$eaf.outcome[cor1]=EAF_num[cor1]
    data_MAF$type[cor1]='corrected'
    cor2<-which(data_MAF$other_allele.outcome ==data_MAF$A1)
    cor21<-setdiff(cor2,cor1)
    cor12<-setdiff(cor1,cor2)
    error<-c(cor12,cor21)
    data_MAF$eaf.outcome[error]=NA
    data_MAF$type[error]='error'
    
    data_MAF<-list(data_MAF=data_MAF,cor1=cor1,harna=harna,error=error)
    
    return(data_MAF)
    
  }
  
  if(type=='exposure' && (is.na(dat$eaf.exposure[1])==T || is.null(dat$eaf.exposure)==T)){
    r<-nrow(dat)
    
    setwd(path)
    MAF<-fread('fileFrequency.frq',header = T)
    
    dat<-merge(dat,MAF,by.x = 'SNP',by.y = 'SNP',all.x = T)
    
    dat<-corrected_eaf_expo(dat)
    
    cor1<-dat$cor1
    
    harna<-dat$harna
    
    error<-dat$error
    
    dat<-dat$data_MAF
    
    print(paste0('一共有',(r-length(harna)-length(error)),'個SNP成功匹配EAF,占比',(r-length(harna)-length(error))/r*100,'%'))
    
    print(paste0('一共有',length(cor1),'個SNP是major allele，EAF被計算為1-MAF,在成功匹配數(shù)目中占比',length(cor1)/(r-length(harna)-length(error))*100,'%'))
    
    print(paste0('一共有',length(harna),'個SNP在1000G中找不到，占比',length(harna)/r*100,'%'))
    
    print(paste0('一共有',length(error),'個SNP在輸入數(shù)據(jù)與1000G中效應(yīng)列與參照列，將剔除eaf，占比',length(error)/r*100,'%'))
    
    print('輸出數(shù)據(jù)中的type列說明：')
    print('raw：EAF直接等于1000G里的MAF數(shù)值，因為效應(yīng)列是minor allele')
    print('corrected：EAF等于1000G中1-MAF，因為效應(yīng)列是major allele')
    print('error：輸入數(shù)據(jù)與1000G里面提供的數(shù)據(jù)完全不一致，比如這個SNP輸入的效應(yīng)列是C，參照列是G，但是1000G提供的是A-T，這種情況下，EAF會被清空（NA），當(dāng)成匹配失敗')
    
    return(dat)
  }
  
  if(type=='outcome' && (is.na(dat$eaf.outcome[1])==T || is.null(dat$eaf.outcome)==T)){
    r<-nrow(dat)
    
    setwd(path)
    MAF<-fread('fileFrequency.frq',header = T)
    
    dat<-merge(dat,MAF,by.x = 'SNP',by.y = 'SNP',all.x = T)
    
    dat<-corrected_eaf_out(dat)
    
    cor1<-dat$cor1
    
    harna<-dat$harna
    
    error<-dat$error
    
    dat<-dat$data_MAF
    
    print(paste0('一共有',(r-length(harna)-length(error)),'個SNP成功匹配EAF,占比',(r-length(harna)-length(error))/r*100,'%'))
    
    print(paste0('一共有',length(cor1),'個SNP是major allele，EAF被計算為1-MAF,在成功匹配數(shù)目中占比',length(cor1)/(r-length(harna)-length(error))*100,'%'))
    
    print(paste0('一共有',length(harna),'個SNP在1000G找不到，占比',length(harna)/r*100,'%'))
    
    print(paste0('一共有',length(error),'個SNP在輸入數(shù)據(jù)與1000G中效應(yīng)列與參照列，將剔除eaf，占比',length(error)/r*100,'%'))
    
    print('輸出數(shù)據(jù)中的type列說明：')
    print('raw：EAF直接等于1000G里的MAF數(shù)值，因為效應(yīng)列是minor allele')
    print('corrected：EAF等于1000G中1-MAF，因為效應(yīng)列是major allele')
    print('error：輸入數(shù)據(jù)與1000G里面提供的數(shù)據(jù)完全不一致，比如這個SNP輸入的效應(yīng)列是C，參照列是G，但是1000G提供的是A-T，這種情況下，EAF會被清空（NA），當(dāng)成匹配失敗')
    
    return(dat)
  }
  else{return(dat)}
}

運行這個函數(shù)需要注意，如果你的數(shù)據(jù)是自己整理的本地數(shù)據(jù)，那就要提前將數(shù)據(jù)整理一下：

exp_dat <- read_exposure_data(
  filename = './Mediators/newcytokines_exposure.txt', ##你的數(shù)據(jù)
  clump = FALSE,
  sep= '\t',
  snp_col = 'rsid',
  beta_col = 'BETA_exposure',
  se_col = 'SE_exposure',
  effect_allele_col ='EA_exposure',
  other_allele_col = 'NEA_exposure',
  eaf_col = 'eaf.exposure',
  pval_col = 'P'
)
head(exp_dat)

# 從1000G的MAF文件中提取EAF并將其與輸入數(shù)據(jù)匹配
newdat <- get_eaf_from_1000G(exp_dat, 'D:/09-MY_try/MR_Aanalyses/', type = 'exposure') ##這里的路徑是包含fileFrequency.frq這個文件的文件夾！