source('https://bioconductor.org/biocLite.R')#加载的网址都一样biocLite("edgeR")#把双引号的内容换成你所需要程序包的名称即可

（express_rec[express_rec==0<-1）

rownames(group_text) <- group_text[,1]group_text <- group_text[c(-1)]Group <- factor(group_text$group,levels = c('Tumor','Normal'))design <- model.matrix(~0+Group)colnames(design) <- c('Tumor','Normal')rownames(design) <- rownames(group_text)

fit <- lmFit(express_rec,design)#制作比对标准；contrast.matrix <- makeContrasts(Tumor - Normal,levels=design)fit2 <- contrasts.fit(fit,contrast.matrix)#进行经验贝叶斯检验；fit2 <- eBayes(fit2)#基于logFC为标准，设置数量上限为30000，调整方法为fdr;all_diff <- topTable(fit2, adjust.method = 'fdr',coef=1,p.value = 1,lfc <- log(1,2),number = 30000,sort.by = 'logFC')#从高到低排名；

#limma的另一种方法；dge <- DGEList(counts = express_rec)dge <- calcNormFactors(dge)#表达矩阵进行标准化；v <- voom(dge, design,plot=TRUE)#利用limma_voom方法进行差异分析；fit <- lmFit(v, design)#对数据进行线性拟合；fit <- eBayes(fit)#贝叶斯算法组建all <- topTable(fit, coef=ncol(design),n=Inf)#从高到低排名；sig.limma <- subset(all_diff,abs(all$logFC)>1.5&all$P.Value<0.05)#进行差异基因筛选；write.csv(sig.limma,'C:/Users/FREEDOM/Desktop/TCGA_data/limm_diff.csv')#写入csv文件中；

express_rec<- read.csv('C:/Users/FREEDOM/Desktop/TCGA_data/after_note2.csv')#读取数据group_text <- read.csv('C:/Users/FREEDOM/Desktop/TCGA_data/group_text.csv')library('DESeq2')#加载包；install.packages('rpart')#含有这个包可忽略，没有的时候才安装；express_rec <- express_rec[,-1]rownames(express_rec) <-express_rec[,1]express_rec <- express_rec[(-1)]#表达矩阵的处理；rownames(group_text) <- group_text[,1]

rownames(group_text) <- group_text[,1]group_text <- group_text[c(-1)]#分组矩阵的数据处理；all(rownames(group_text)==colnames(express_rec))#确保表达矩阵的列名与分组矩阵行名相一致；

dds <- DESeqDataSetFromMatrix(countData=express_rec, colData=group_text, design<- ~ group)  #DESeq2的加载head(dds)dds <- dds[rowSums(counts(dds)) > 1, ] #过滤一些low count的数据；

> dds <- DESeq(dds)#DESeq进行标准化；estimating size factorsestimating dispersionsgene-wise dispersion estimatesmean-dispersion relationshipfinal dispersion estimatesfitting model and testing-- replacing outliers and refitting for 2819 genes-- DESeq argument 'minReplicatesForReplace' = 7 -- original counts are preserved in counts(dds)estimating dispersionsfitting model and testing

> summary(res)#查看经过标准化矩阵的基本情况；

out of 24823 with nonzero total read countadjusted p-value < 0.1LFC > 0 (up)       : 7590, 31%LFC < 0 (down)     : 5120, 21%outliers [1]       : 0, 0%low counts [2]     : 0, 0%(mean count < 0)[1] see 'cooksCutoff' argument of ?results[2] see 'independentFiltering' argument of ?results

path = 'C:/Users/FREEDOM/Desktop/TCGA_data/after_note2.csv'path1 ='C:/Users/FREEDOM/Desktop/TCGA_data/group_text.csv'express_rec <- read.csv(path,headers <- T)#读取表达矩阵;group_text <- read.csv(path1,headers <- T)#读取分组矩阵；library(edgeR)#加载edgeR包express_rec <- express_rec[,-1]rownames(express_rec) <- express_rec[,1]express_rec <- express_rec[(-1)]#创建表达矩阵；rownames(group_text) <- group_text[,1]group_text <- group_text[c(-1)]#加载分组矩阵；group <-factor(group_text$group)dge <- DGEList(counts = express_rec,group = group)#构建DEList对象；y <- calcNormFactors(dge)#利用calcNormFactor函数对DEList对象进行标准化(TMM算法) #创建设计矩阵，跟Limma包相似；Group <- factor(group_text$group,levels = c('Tumor','Normal'))design <- model.matrix(~0+Group)colnames(design) <- c('Tumor','Normal')rownames(design) <- rownames(group_text)#创建分组矩阵；

dge <- DGEList(counts = express_rec,group = group)#构建DEList对象；y <- calcNormFactors(dge)#利用calcNormFactor函数对DEList对象进行标准化(TMM算法) y <- estimateDisp(y,design)#估计离散值（Dispersion）fit <- glmQLFit(y, design, robust=TRUE)#进一步通过quasi-likelihood (QL)拟合NB模型TU.vs.NO <- makeContrasts(Tumor-Normal, levels=design)#这一步主要构建比较矩阵；res <- glmQLFTest(fit, contrast=TU.vs.NO)#用QL F-test进行检验# ig.edger <- res$table[p.adjust(res$table$PValue, method = "BH") < 0.01, ]#利用‘BH’方法；result_diff <- res$table#取出最终的差异基因；write.csv(edge_diff,'C:/Users/FREEDOM/Desktop/TCGA_data/edgeR_diff2.csv')