RNA-seq(7): DEseq2筛选差异表达基因并注释(bi
2018-08-02 本文已影响38人
Y大宽
序
- 走到这一步，似乎顺畅了许多，最主要时间不用花费那么多了。另外，以前曾经处理过不计其数的芯片，挑选差异表达基因，筛选关键基因，功能富集，还有基于全部数据的WGCNA(当然你也可以用差异基因来做，虽然不推荐，看不少文章也这么发），GSEA，PPI等等，这些后续我会慢慢发出来。
- 但是，因为以前处理的芯片表达谱数据是符合正态分布，所以可以用t检验来筛选差异表达基因，但RNA-seq的read count普遍认为符合泊松分布。所以筛选DEGs的方法还是不一样
------------要求---------------
- 载入表达矩阵
- 设置好分组信息
- 用DEseq2进行差异分析
- 输出差异分析结果
来源于生信技能树
-------------------------参考文章-----------------------------------
- 2018.7月:Analyzing RNA-seq data with DESeq2
- Count-Based Differential Expression Analysis of RNA-seq Data
- 强烈推荐：Tutorial: RNA-seq differential expression & pathway analysis with Sailfish, DESeq2, GAGE, and Pathview
- 可以下载的pdf havard,点此下载
-------------------开始------------------------
-
正式载入数据之前，先看一下数据结构data structure，因为这是我们下面一切分析的起点。
-
我们需要准备2个table：一个是countData，一个是colData
two kinds of data.png
关于上面两个表的说明
-
countData表示的是count矩阵，行代表gene，列代表样品，中间的数字代表对应count数。colData表示sample的元数据，因为这个表提供了sample的元数据。
because this table supplies metadata/information about the columns of the countData matrix. Notice that the first column of colData must match the column names of countData (except the first, which is the gene ID column). - colData的行名与countData的列名一致（除去代表gene ID的第一列）
1 载入数据（countData和colData）
> library(tidyverse)
> library(DESeq2)
> #import data
> setwd("F:/rna_seq/data/matrix")
> mycounts<-read.csv("readcount.csv")
> head(mycounts)
X control1 control2 treat1 treat2
1 ENSMUSG00000000001 1648 2306 2941 2780
2 ENSMUSG00000000003 0 0 0 0
3 ENSMUSG00000000028 835 950 1366 1051
4 ENSMUSG00000000031 65 83 52 53
5 ENSMUSG00000000037 70 53 94 66
6 ENSMUSG00000000049 0 3 4 5
#这里有个x，需要去除，先把第一列当作行名来处理
> rownames(mycounts)<-mycounts[,1]
#把带X的列删除
> mycounts<-mycounts[,-1]
> head(mycounts)
control1 control2 treat1 treat2
ENSMUSG00000000001 1648 2306 2941 2780
ENSMUSG00000000003 0 0 0 0
ENSMUSG00000000028 835 950 1366 1051
ENSMUSG00000000031 65 83 52 53
ENSMUSG00000000037 70 53 94 66
ENSMUSG00000000049 0 3 4 5
# 这一步很关键，要明白condition这里是因子，不是样本名称；小鼠数据有对照组和处理组，各两个重复
> condition <- factor(c(rep("control",2),rep("treat",2)), levels = c("control","treat"))
> condition
[1] control control treat treat
Levels: control treat
#colData也可以自己在excel做好另存为.csv格式，再导入即可
> colData <- data.frame(row.names=colnames(mycounts), condition)
> colData
condition
control1 control
control2 control
treat1 treat
treat2 treat
2构建dds对象,开始DESeq流程
注释：dds=DESeqDataSet Object
> dds <- DESeqDataSetFromMatrix(mycounts, colData, design= ~ condition)
> dds <- DESeq(dds)
> # 查看一下dds的内容
> dds
显示为
class: DESeqDataSet
dim: 6 4
metadata(1): version
assays(3): counts mu cooks
rownames(6): ENSMUSG00000000001 ENSMUSG00000000003 ... ENSMUSG00000000037 ENSMUSG00000000049
rowData names(21): baseMean baseVar ... deviance maxCooks
colnames(4): control1 control2 treat1 treat2
colData names(2): condition sizeFactor
3 总体结果查看
接下来，我们要查看treat versus control的总体结果，并根据p-value进行重新排序。利用summary命令统计显示一共多少个genes上调和下调（FDR0.1）
> res = results(dds, contrast=c("condition", "control", "treat"))
#或下面命令
> res= results(dds)
> res = res[order(res$pvalue),]
> head(res)
> summary(res)
#所有结果先进行输出
> write.csv(res,file="All_results.csv")
> table(res$padj<0.05)
上述代码的结果显示
> res = results(dds2, contrast=c("condition", "control", "treat"))
> res = res[order(res$pvalue),]
> head(res)
log2 fold change (MLE): condition control vs treat
Wald test p-value: condition control vs treat
DataFrame with 6 rows and 6 columns
baseMean log2FoldChange lfcSE stat pvalue padj
<numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
ENSMUSG00000003309 548.1926 3.231611 0.2658125 12.157485 5.234568e-34 8.193146e-30
ENSMUSG00000046323 404.1894 3.067050 0.2628220 11.669687 1.820923e-31 1.425055e-27
ENSMUSG00000001123 341.8542 2.797485 0.2766499 10.112004 4.887441e-24 2.549941e-20
ENSMUSG00000023906 951.9460 2.382307 0.2510718 9.488551 2.342684e-21 9.116395e-18
ENSMUSG00000018569 485.4839 3.136031 0.3312999 9.465836 2.912214e-21 9.116395e-18
ENSMUSG00000000184 601.0842 -2.827750 0.3154171 -8.965112 3.099648e-19 8.085948e-16
> summary(res)
out of 28335 with nonzero total read count
adjusted p-value < 0.1
LFC > 0 (up) : 445, 1.6%
LFC < 0 (down) : 625, 2.2%
outliers [1] : 0, 0%
low counts [2] : 12683, 45%
(mean count < 18)
[1] see 'cooksCutoff' argument of ?results
[2] see 'independentFiltering' argument of ?results
> write.csv(res,file="All_results.csv")
> table(res$padj<0.05)
FALSE TRUE
14909 743
可见，一共445个genes上调，625个genes下调，没有离群值。padj小于0.05的共有743个。
4 提取差异表达genes（DEGs）并进行gene symbol注释
差异表达基因的界定很不统一，但log2FC是用的最广泛同时也是最不精确的方式，但因为其好理解所以广泛被应用尤其芯片数据处理中，记的是havard universit做过一个统计，FC=2相对比较可靠。但无论怎么，这种界定人为因素太大，过于武断。所以GSEA，WGCNA是拿全部表达数据(可以进行初步过滤)来进行分析，并且WGCNA采取软阈值的方式来挑选genes更加合理。既然挑选差异表达基因，还是采取log2FC和padj来进行。
获取padj（p值经过多重校验校正后的值）小于0.05，表达倍数取以2为对数后大于1或者小于-1的差异表达基因。代码如下
> diff_gene_deseq2 <-subset(res, padj < 0.05 & abs(log2FoldChange) > 1)
#或
#> diff_gene_deseq2 <-subset(res,padj < 0.05 & (log2FoldChange > 1 | log2FoldChange < -1))
> dim(diff_gene_deseq2)
> head(diff_gene_deseq2)
> write.csv(diff_gene_deseq2,file= "DEG_treat_vs_control.csv")
结果显示如下：
> diff_gene_deseq2 <-subset(res, padj < 0.05 & abs(log2FoldChange) > 1)
> dim(diff_gene_deseq2)
[1] 431 6
> head(diff_gene_deseq2)
log2 fold change (MLE): condition control vs treat
Wald test p-value: condition control vs treat
DataFrame with 6 rows and 6 columns
baseMean log2FoldChange lfcSE stat pvalue padj
<numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
ENSMUSG00000003309 548.1926 3.231611 0.2658125 12.157485 5.234568e-34 8.193146e-30
ENSMUSG00000046323 404.1894 3.067050 0.2628220 11.669687 1.820923e-31 1.425055e-27
ENSMUSG00000001123 341.8542 2.797485 0.2766499 10.112004 4.887441e-24 2.549941e-20
ENSMUSG00000023906 951.9460 2.382307 0.2510718 9.488551 2.342684e-21 9.116395e-18
ENSMUSG00000018569 485.4839 3.136031 0.3312999 9.465836 2.912214e-21 9.116395e-18
ENSMUSG00000000184 601.0842 -2.827750 0.3154171 -8.965112 3.099648e-19 8.085948e-16
5 用bioMart对差异表达基因进行注释
和RNA-seq(6): reads计数，合并矩阵并进行注释代码一样
library('biomaRt')
library("curl")
mart <- useDataset("mmusculus_gene_ensembl", useMart("ensembl"))
my_ensembl_gene_id<-row.names(diff_gene_deseq2)
#listAttributes(mart)
mms_symbols<- getBM(attributes=c('ensembl_gene_id','external_gene_name',"description"),
filters = 'ensembl_gene_id', values = my_ensembl_gene_id, mart = mart)
结果如下：
> library('biomaRt')
> library("curl")
> mart <- useDataset("mmusculus_gene_ensembl", useMart("ensembl"))
> my_ensembl_gene_id<-row.names(diff_gene_deseq2)
> mms_symbols<- getBM(attributes=c('ensembl_gene_id','external_gene_name',"description"),
+ filters = 'ensembl_gene_id', values = my_ensembl_gene_id, mart = mart)
> head(mms_symbols)
ensembl_gene_id external_gene_name description
1 ENSMUSG00000000120 Ngfr nerve growth factor receptor (TNFR superfamily, member 16) [Source:MGI Symbol;Acc:MGI:97323]
2 ENSMUSG00000000184 Ccnd2 cyclin D2 [Source:MGI Symbol;Acc:MGI:88314]
3 ENSMUSG00000000276 Dgke diacylglycerol kinase, epsilon [Source:MGI Symbol;Acc:MGI:1889276]
4 ENSMUSG00000000308 Ckmt1 creatine kinase, mitochondrial 1, ubiquitous [Source:MGI Symbol;Acc:MGI:99441]
5 ENSMUSG00000000320 Alox12 arachidonate 12-lipoxygenase [Source:MGI Symbol;Acc:MGI:87998]
6 ENSMUSG00000000708 Kat2b K(lysine) acetyltransferase 2B [Source:MGI Symbol;Acc:MGI:1343094]
5合并数据:res结果+mms_symbols合并成一个文件
合并的话两个数据必须有共同的列名，我们先看一下
> head(diff_gene_deseq2)
log2 fold change (MLE): condition control vs treat
Wald test p-value: condition control vs treat
DataFrame with 6 rows and 6 columns
baseMean log2FoldChange lfcSE stat pvalue padj
<numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
ENSMUSG00000003309 548.1926 3.231611 0.2658125 12.157485 5.234568e-34 8.193146e-30
ENSMUSG00000046323 404.1894 3.067050 0.2628220 11.669687 1.820923e-31 1.425055e-27
ENSMUSG00000001123 341.8542 2.797485 0.2766499 10.112004 4.887441e-24 2.549941e-20
ENSMUSG00000023906 951.9460 2.382307 0.2510718 9.488551 2.342684e-21 9.116395e-18
ENSMUSG00000018569 485.4839 3.136031 0.3312999 9.465836 2.912214e-21 9.116395e-18
ENSMUSG00000000184 601.0842 -2.827750 0.3154171 -8.965112 3.099648e-19 8.085948e-16
> head(mms_symbols)
ensembl_gene_id external_gene_name
1 ENSMUSG00000000120 Ngfr
2 ENSMUSG00000000184 Ccnd2
3 ENSMUSG00000000276 Dgke
4 ENSMUSG00000000308 Ckmt1
5 ENSMUSG00000000320 Alox12
6 ENSMUSG00000000708 Kat2b
可见，两个文件没有共同的列名，所以要先给'diff_gene_deseq2'添加一个‘ensembl_gene_id’的列名，方法如下:(应该有更简便的方法)
> ensembl_gene_id<-rownames(diff_gene_deseq2)
> diff_gene_deseq2<-cbind(ensembl_gene_id,diff_gene_deseq2)
> colnames(diff_gene_deseq2)[1]<-c("ensembl_gene_id")
> diff_name<-merge(diff_gene_deseq2,mms_symbols,by="ensembl_gene_id")
>head(diff_name)
#查看Akap8的情况
Akap8 <- diff_name[diff_name$external_gene_name=="Akap8",]
中间显示过程如下：
> ensembl_gene_id<-rownames(diff_gene_deseq2)
> diff_gene_deseq2<-cbind(ensembl_gene_id,diff_gene_deseq2)
> colnames(diff_gene_deseq2)[1]<-c("ensembl_gene_id")
> diff_name<-merge(diff_gene_deseq2,mms_symbols,by="ensembl_gene_id")
> head(diff_name)
DataFrame with 6 rows and 9 columns
ensembl_gene_id baseMean log2FoldChange lfcSE stat pvalue padj external_gene_name
<character> <numeric> <numeric> <numeric> <numeric> <numeric> <numeric> <character>
1 ENSMUSG00000000120 434.04177 -1.293648 0.2713146 -4.768072 1.859970e-06 2.064698e-04 Ngfr
2 ENSMUSG00000000184 601.08417 -2.827750 0.3154171 -8.965112 3.099648e-19 8.085948e-16 Ccnd2
3 ENSMUSG00000000276 668.12500 -1.071362 0.2445381 -4.381168 1.180446e-05 9.603578e-04 Dgke
4 ENSMUSG00000000308 207.46719 1.944949 0.3427531 5.674489 1.391035e-08 3.819733e-06 Ckmt1
5 ENSMUSG00000000320 61.96266 1.451927 0.4637101 3.131109 1.741473e-03 4.105051e-02 Alox12
6 ENSMUSG00000000708 1070.03203 -1.046546 0.2049062 -5.107440 3.265530e-07 5.056107e-05 Kat2b
description
<character>
1 nerve growth factor receptor (TNFR superfamily, member 16) [Source:MGI Symbol;Acc:MGI:97323]
2 cyclin D2 [Source:MGI Symbol;Acc:MGI:88314]
3 diacylglycerol kinase, epsilon [Source:MGI Symbol;Acc:MGI:1889276]
4 creatine kinase, mitochondrial 1, ubiquitous [Source:MGI Symbol;Acc:MGI:99441]
5 arachidonate 12-lipoxygenase [Source:MGI Symbol;Acc:MGI:87998]
6 K(lysine) acetyltransferase 2B [Source:MGI Symbol;Acc:MGI:1343094]
> Akap8 <- diff_name[diff_name$external_gene_name=="Akap8",]
> Akap8
DataFrame with 1 row and 9 columns
ensembl_gene_id baseMean log2FoldChange lfcSE stat pvalue padj external_gene_name
<character> <numeric> <numeric> <numeric> <numeric> <numeric> <numeric> <character>
1 ENSMUSG00000024045 2281.053 -1.276329 0.2428315 -5.256028 1.471996e-07 2.775865e-05 Akap8
description
<character>
1 A kinase (PRKA) anchor protein 8 [Source:MGI Symbol;Acc:MGI:1928488]
至此，差异表达基因提取并注释完毕，下一步
-
先进行数据可视化(Data visulization)
-
然后进行富集分分析及可视化
