基因的差异表达分析，通常使用R中的软件包，包括：DESeq2，edgeR，limma等，今天介绍DESeq2的分析流程：
1、在R中安装DESeq2软件包
source ("http://bioconductor.org/biocLite.R") # 调入安装工具Bioconductor
biocLite("DESeq2") # 安装DESeq2
library("DESeq2") # 测试是否安装成功
如果安装出错，请移步前文：《MAC安装DESeq2报错及解决方案》，作为一个生信小白，我花了3天时间才把安装DESeq2的问题解决，一路全是坑啊！
2、导入分析数据
我们还是使用《RNASEQ分析入门笔记7-HTSeq定量基因表达水平》中得到的数据进行分析，首先回顾一下：
上文中，我们得到了raw_count_filt2矩阵，格式如下（执行：tail (raw_count_filt2, n=5):
| control1 | control2 | rep1 | rep2 | |
|---|---|---|---|---|
| ENSMUSG00000110420 | 0 | 0 | 0 | 0 |
| ENSMUSG00000110421 | 0 | 0 | 0 | 1 |
| ENSMUSG00000110422 | 1 | 0 | 1 | 2 |
| ENSMUSG00000110423 | 0 | 0 | 0 | 0 |
| ENSMUSG00000110424 | 26 | 20 | 48 | 24 |
可以看到，该矩阵实际只有4列，是由整数组成的，而最前面的一列是行名，可以直接从保存的.csv文件中导入：
raw_count_filt2 <- read.csv ("raw_count_filt2.csv") # 从保存的.csv文件导入数据，并赋值给raw_count_filt2
3、加载DESeq2并设置样品信息
library(DESeq2) # 加载DESeq2包
countData <- raw_count_filt2 # 表达矩阵
condition <- factor(c("control","control","KD","KD")) # 定义condition
colData <- data.frame(row.names=colnames(countData), condition) # 样品信息矩阵
condition：
[1] control control KD KD
Levels: KD control
colData：
| condition | |
|---|---|
| control1 | control |
| control2 | control |
| rep1 | KD |
| rep2 | KD |
4、构建dds矩阵
dds <- DESeqDataSetFromMatrix(countData, DataFrame(condition), design= ~ condition ) # 构建dds矩阵
head(dds) # 查看dds矩阵的前6行
class: DESeqDataSet
dim: 6 4
metadata(1): version
assays(1): counts
rownames(6): ENSMUSG00000000001 ENSMUSG00000000003 ...
ENSMUSG00000000037 ENSMUSG00000000049
rowData names(0):
colnames(4): control1 control2 rep1 rep2
colData names(1): condition
dds2 <- DESeq(dds) # 对dds进行Normalize
运行成功会有如下提示：
estimating size factors
estimating dispersions
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates
fitting model and testing
resultsNames（dds) # 查看结果的名称
[1] "Intercept" "condition_control_vs_KD"
res <- results(dds) # 使用results()函数获取结果，并赋值给res
head (res, n=5) # 查看res矩阵的前5行
输出的res矩阵有6列，分别是：baseMean，log2FoldChange，lfcSE，pvalue，padj：
log2 fold change (MLE): condition control vs KD
Wald test p-value: condition control vs KD
DataFrame with 5 rows and 6 columns
| baseMean | log2FoldChange | lfcSE | stat | pvalue | padj | |
|---|---|---|---|---|---|---|
| ENSMUSG00000000001 | 2359.75391 | -0.08461265 | 0.2081281 | -0.4065413 | 0.68434494 | 0.9040654 |
| ENSMUSG00000000003 | 0.00000 | NA | NA | NA | NA | NA |
| ENSMUSG00000000028 | 1027.82363 | 0.03638541 | 0.2036829 | 0.1786375 | 0.85822232 | 0.9618582 |
| ENSMUSG00000000031 | 65.37333 | 0.94588315 | 0.4428518 | 2.1358912 | 0.03268828 | 0.2511131 |
| ENSMUSG00000000037 | 69.75843 | 0.11587980 | 0.4400431 | 0.2633374 | 0.79229054 | 0.9422796 |
mcols(res,use.names= TRUE) # 查看res矩阵每一列的含义
DataFrame with 6 rows and 2 columns
| type | description | |
|---|---|---|
| baseMean | intermediate | mean of normalized counts for all samples |
| log2FoldChange | results | log2 fold change (MLE): condition control vs KD |
| lfcSE | results | standard error: condition control vs KD |
| stat | results | Wald statistic: condition control vs KD |
| pvalue | results | Wald test p-value: condition control vs KD |
| padj | results | BH adjusted p-values |
summary(res) # 对res矩阵进行总结
out of 28335 with nonzero total read count
adjusted p-value < 0.1
LFC > 0 (up) : 445, 1.6%
LFC < 0 (down) : 625, 2.2%
outliers [1] : 0, 0%
low counts [2] : 12683, 45%
(mean count < 18)
[1] see 'cooksCutoff' argument of ?results
[2] see 'independentFiltering' argument of ?results
其中，445个基因表达上调，625个基因表达下降，可靠程度是：p-value < 0.1
5、提取差异分析结果
table(res$padj<0.05) # 取padj小于0.05的数据，得到743行
| FALSE | TRUE |
|---|---|
| 14909 | 743 |
res <- res[order(res$padj),] # 按照padj的大小将res重新排列
diff_gene_deseq2 <- subset(res,padj < 0.05 & (log2FoldChange >1 | log2FoldChange < -1)) # 获取padj小于0.05，表达倍数取以2为对数后绝对值大于1的差异表达基因，赋值给diff_gene_deseq2
head (diff_gene_deseq2, n=5) # 查看diff_gene_deseq2矩阵的前5行
log2 fold change (MLE): condition control vs KD
Wald test p-value: condition control vs KD
DataFrame with 5 rows and 6 columns
| baseMean | log2FoldChange | lfcSE | stat | pvalue | padj | |
|---|---|---|---|---|---|---|
| ENSMUSG00000003309 | 548.1926 | 3.231612 | 0.2658125 | 12.157487 | 5.234431e-34 | 8.192931e-30 |
| ENSMUSG00000046323 | 404.1894 | 3.067051 | 0.2628220 | 11.669689 | 1.820880e-31 | 1.425021e-27 |
| ENSMUSG00000001123 | 341.8542 | 2.797485 | 0.2766499 | 10.112006 | 4.887336e-24 | 2.549886e-20 |
| ENSMUSG00000018569 | 485.4839 | 3.136032 | 0.3312999 | 9.465839 | 2.912140e-21 | 9.116163e-18 |
| ENSMUSG00000023906 | 951.9460 | 2.382308 | 0.2510718 | 9.488553 | 2.342631e-21 | 9.116163e-18 |
diff_gene_deseq2 <- row.names(diff_gene_deseq2) # 提取diff_gene_deseq2的行名
head (diff_gene_deseq2, n=5)
[1] "ENSMUSG00000003309"
[2] "ENSMUSG00000046323"
[3] "ENSMUSG00000001123"
[4] "ENSMUSG00000018569"
[5] "ENSMUSG00000023906"
resdata <- merge (as.data.frame(res),as.data.frame(counts(dds,normalize=TRUE)),by="row.names",sort=FALSE)
head (resdata,n=5)
| Row.names | baseMean | log2FoldChange | lfcSE | stat | pvalue | padj | control1 | control2 | rep1 | rep2 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | ENSMUSG00000003309 | 548.1926 | 3.231612 | 0.2658125 | 12.157487 | 5.234431e-34 | 8.192931e-30 | 1073.9977 | 908.7239 | 125.93083 | 84.11803 |
| 2 | ENSMUSG00000046323 | 404.1894 | 3.067051 | 0.2628220 | 11.669689 | 1.820880e-31 | 1.425021e-27 | 723.4828 | 721.5049 | 97.10329 | 74.66657 |
| 3 | ENSMUSG00000001123 | 341.8542 | 2.797485 | 0.2766499 | 10.112006 | 4.887336e-24 | 2.549886e-20 | 679.8243 | 515.6735 | 84.96538 | 86.95347 |
| 4 | ENSMUSG00000018569 | 485.4839 | 3.136032 | 0.3312999 | 9.465839 | 2.912140e-21 | 9.116163e-18 | 1113.9140 | 630.6325 | 127.44807 | 69.94083 |
| 5 | ENSMUSG00000023906 | 951.9460 | 2.382308 | 0.2510718 | 9.488553 | 2.342631e-21 | 9.116163e-18 | 1862.3445 | 1333.5250 | 351.99943 | 259.91526 |
write.csv(resdata, file="control_vs_akap95.csv") # 将结果写入control_vs_akap95.csv文件