在甲基化ChAMP包中加入协变量

最近在分析甲基化数据，发现ChAMP包的输入端是beta（甲基化数据）和pheno（表型数据），找不到校正协变量（比如年龄、性别、批次等）的输入端。

如下代码所示：

champ.DMP(beta = myNorm,
              pheno = myLoad$pd$Sample_Group,
              compare.group = NULL,
              adjPVal = 0.05,
              adjust.method = "BH",
              arraytype = "450K")

查了一下ChAMP包的代码，发现ChAMP包调用的是limma函数。

因此，这事就比较好解决了。

只需要在原代码的基础上修改model就行了。

下面介绍一下如何修改model。

1创建甲基化数据集和表型、协变量数据集

beta <- matrix(runif(60,min=0,max=1),10,6)
rownames(beta) <- c("cg18478105","cg09835024","cg14361672","cg01763666","cg12950382","cg02115394","cg25813447","cg07779434","cg13417420","cg12480843")
colnames(beta) <-paste("sample",1:6)
da=data.frame(pheno=c("1","1","1","2","2","2"),SEX=c("1","2","1","1","1","2"),AGE=c(54,23,58,43,68,36))
rownames(da) <- paste("sample",1:6)

甲基化数据如下所示：

注意：beta数据是随机产生的，因此每个人生成的数据是不一样的。

表型（pheno）、协变量（SEX、AGE）的数据如下：

2 修改model、校正协变量

在这里我们假定要校正的协变量有SEX和AGE。

我在tian yuan创作的champ.DMP函数的基础上增加了cov1=da$cov1和cov2=da$cov2两个输入参数。
在champ.DMP函数里面修改了

design <- model.matrix( ~ 0+ factor(p)+cov1+cov2 )
colnames(design) <- c("control", "case","cov1","cov2")
contrast.matrix <- makeContrasts(contrasts=paste(colnames(design)[2:1],collapse="-"), levels = design)

最后修改完的函数如下所示：

champ.DMP1 <- function(beta = myNorm,
                      pheno = da$Sample_Group,
                      cov1=da$cov1,
                      cov2=da$cov2,
                      adjPVal = 0.05,
                      adjust.method = "BH",
                      compare.group = NULL,
                      arraytype = "EPIC")
{
    message("[===========================]")
    message("[<<<<< ChAMP.DMP START >>>>>]")
    message("-----------------------------")

    if(is.null(pheno) | length(unique(pheno))<=1)
    {
        stop("pheno parameter is invalid. Please check the input, pheno MUST contain at least two phenotypes.")
    }else
    {
        message("<< Your pheno information contains following groups. >>")
        sapply(unique(pheno),function(x) message("<",x,">:",sum(pheno==x)," samples."))
        message("[The power of statistics analysis on groups contain very few samples may not strong.]")
    }
	
    if(is.null(compare.group))
    {
        message("You did not assign compare groups. The first two groups: <",unique(pheno)[1],"> and <",unique(pheno)[2],">, will be compared automatically.")
        compare.group <- unique(pheno)[1:2]
    }else if(sum(compare.group %in% unique(pheno))==2)
    {
        message("As you assigned, champ.DMP will compare ",compare.group[1]," and ",compare.group[2],".")
    }else
    {
        message("Seems you did not assign correst compare groups. The first two groups: <",unique(pheno)[1],"> and <",unique(pheno)[2],">, will be compared automatically.")
        compare.group <- unique(pheno)[1:2]
    }
    
    p <- pheno[which(pheno %in% compare.group)]
    beta <- beta[,which(pheno %in% compare.group)]
   design <- model.matrix( ~ 0+ factor(p)+cov1+cov2 )
	colnames(design) <- c("control", "case","cov1","cov2")
    contrast.matrix <- makeContrasts(contrasts=paste(colnames(design)[2:1],collapse="-"), levels = design)
    message("
<< Contrast Matrix >>")
    print(contrast.matrix)

    message("
<< All beta, pheno and model are prepared successfully. >>")
	
	fit <- lmFit(beta, design)
	fit2 <- contrasts.fit(fit,contrast.matrix)
	tryCatch(fit3 <- eBayes(fit2),
      warning=function(w) 
      {
      	stop("limma failed, No sample variance.
")
      })
    DMP <- topTable(fit3,coef=1,number=nrow(beta),adjust.method=adjust.method)
    message("You have found ",sum(DMP$adj.P.Val <= adjPVal), " significant MVPs with a ",adjust.method," adjusted P-value below ", adjPVal,".")
    message("
<< Calculate DMP successfully. >>")

    if(arraytype == "EPIC") data(probe.features.epic) else data(probe.features)
    com.idx <- intersect(rownames(DMP),rownames(probe.features))
    avg <-  cbind(rowMeans(beta[com.idx,which(p==compare.group[1])]),rowMeans(beta[com.idx,which(p==compare.group[2])]))
    avg <- cbind(avg,avg[,2]-avg[,1])
    colnames(avg) <- c(paste(compare.group,"AVG",sep="_"),"deltaBeta")
    DMP <- data.frame(DMP[com.idx,],avg,probe.features[com.idx,])

    message("[<<<<<< ChAMP.DMP END >>>>>>]")
    message("[===========================]")
    message("[You may want to process DMP.GUI() or champ.GSEA() next.]
")
    return(DMP)
}

3、使用champ.DMP1分析甲基化数据

champ.DMP1 <- champ.DMP1(beta =beta,pheno = da$pheno, cov1=da$SEX,cov2=da$AGE, adjust.method = "BH", arraytype = "EPIC")

beta是第一步生成的甲基化数据，da数据框上的pheno、SEX、AGE也是在第一步生成的数据；
arraytype = "EPIC"表示甲基化数据类型为850K。如果是"450K"数据则将"EPIC"改为"450K"；

4、修改技巧

上面举得例子是校正两个协变量，如果想校正三个协变量的话则做如下三个改动：

第一个改动：

champ.DMP1 <- function(beta = myNorm,
                      pheno = da$Sample_Group,
                      cov1=da$cov1,
                      cov2=da$cov2,
                      cov3=da$cov3,
                      adjPVal = 0.05,
                      adjust.method = "BH",
                      compare.group = NULL,
                      arraytype = "EPIC")

第二个改动：

design <- model.matrix( ~ 0+ factor(p)+cov1+cov2+cov3 )
colnames(design) <- c("control", "case","cov1","cov2","cov3")
contrast.matrix <- makeContrasts(contrasts=paste(colnames(design)[2:1],collapse="-"), levels = design)

第三个改动：

champ.DMP1 <- champ.DMP1(beta =beta,pheno = da$pheno, cov1=da$SEX,cov2=da$AGE, cov3=da$cov3,adjust.method = "BH", arraytype = "EPIC")

5、致谢

感谢原作者tian yuan（这个包很全能！）;

感谢健明分享的甲基化分析入门练习：甲基化芯片的一般分析流程

建议各位刚入门甲基化的同学们可以看看健明在B站的视频，讲的很详细。