本文讲的是aglient芯片原始数据的处理，参考资料是limma 的userguide文档。GEO数据库下载的表达矩阵不符合预期，比如是空的，或者是有负值的，那我们就处理一下它的原始数据。aglient的芯片应用也很广泛，举个OSCC的栗子：GSE23558，跟着学习学习。
1.下载和读取数据
1.1获取临床信息数据
从前，提到GEO数据下载，我们只有GEOquery，神功盖世，但是死于网速。后来就有了中国人寄几的GEO镜像，AnnoProbe包。还没有正式发表，就已经初露锋芒了，因简单易学，下载迅速，在我们的粉丝圈子里很受欢迎。
rm(list = ls())
library(stringr)
library(AnnoProbe)
library(GEOquery)
library(limma)
gse = "GSE23558"
geoChina(gse)
## you can also use getGEO from GEOquery, by
## getGEO("GSE23558", destdir=".", AnnotGPL = F, getGPL = F)
## $GSE23558_series_matrix.txt.gz
## ExpressionSet (storageMode: lockedEnvironment)
## assayData: 41000 features, 32 samples
## element names: exprs
## protocolData: none
## phenoData
## sampleNames: GSM577914 GSM577915 ... GSM577945 (32 total)
## varLabels: title geo_accession ... tissue:ch1 (42 total)
## varMetadata: labelDescription
## featureData: none
## experimentData: use 'experimentData(object)'
## pubMedIds: 22072328
## 28433800
## Annotation: GPL6480
提供一个GSE编号就可以下载啦。因为表达矩阵是处理过的，我们不要，所以只提取临床信息表格，从中获得分组信息。
load("GSE23558_eSet.Rdata")
pd <- pData(gset[[1]])
调整pd的行名与文件读取的顺序一致，并定义分组信息。
raw_dir = "rawdata/GSE23558_RAW/"
raw_datas = paste0(raw_dir,"/",dir(raw_dir))
#调整pd与rawdata的顺序一致
raw_order = str_extract(raw_datas,"GSM\\d*")
pd = pd[match(raw_order,rownames(pd)),]
group_list <- ifelse(stringr::str_detect(pd$title,"HealthyControl"),"normal","tumor")
group_list <- factor(group_list, levels=c("normal","tumor"))
1.2 读取原始数据
x <- read.maimages(raw_datas,
source="agilent",
green.only=TRUE,
other.columns = "gIsWellAboveBG")
## Read rawdata/GSE23558_RAW//GSM577914.txt
## Read rawdata/GSE23558_RAW//GSM577915.txt
## Read rawdata/GSE23558_RAW//GSM577916.txt
## Read rawdata/GSE23558_RAW//GSM577917.txt
## Read rawdata/GSE23558_RAW//GSM577918.txt
## Read rawdata/GSE23558_RAW//GSM577919.txt
## Read rawdata/GSE23558_RAW//GSM577920.txt
## Read rawdata/GSE23558_RAW//GSM577921.txt
## Read rawdata/GSE23558_RAW//GSM577922.txt
## Read rawdata/GSE23558_RAW//GSM577923.txt
## Read rawdata/GSE23558_RAW//GSM577924.txt
## Read rawdata/GSE23558_RAW//GSM577925.txt
## Read rawdata/GSE23558_RAW//GSM577926.txt
## Read rawdata/GSE23558_RAW//GSM577927.txt
## Read rawdata/GSE23558_RAW//GSM577928.txt
## Read rawdata/GSE23558_RAW//GSM577929.txt
## Read rawdata/GSE23558_RAW//GSM577930.txt
## Read rawdata/GSE23558_RAW//GSM577931.txt
## Read rawdata/GSE23558_RAW//GSM577932.txt
## Read rawdata/GSE23558_RAW//GSM577933.txt
## Read rawdata/GSE23558_RAW//GSM577934.txt
## Read rawdata/GSE23558_RAW//GSM577935.txt
## Read rawdata/GSE23558_RAW//GSM577936.txt
## Read rawdata/GSE23558_RAW//GSM577937.txt
## Read rawdata/GSE23558_RAW//GSM577938.txt
## Read rawdata/GSE23558_RAW//GSM577939.txt
## Read rawdata/GSE23558_RAW//GSM577940.txt
## Read rawdata/GSE23558_RAW//GSM577941.txt
## Read rawdata/GSE23558_RAW//GSM577942.txt
## Read rawdata/GSE23558_RAW//GSM577943.txt
## Read rawdata/GSE23558_RAW//GSM577944.txt
## Read rawdata/GSE23558_RAW//GSM577945.txt
dim(x)
## [1] 45015 32
2.背景校正和标准化
y <- backgroundCorrect(x, method="normexp")
## Array 1 corrected
## Array 2 corrected
## Array 3 corrected
## Array 4 corrected
## Array 5 corrected
## Array 6 corrected
## Array 7 corrected
## Array 8 corrected
## Array 9 corrected
## Array 10 corrected
## Array 11 corrected
## Array 12 corrected
## Array 13 corrected
## Array 14 corrected
## Array 15 corrected
## Array 16 corrected
## Array 17 corrected
## Array 18 corrected
## Array 19 corrected
## Array 20 corrected
## Array 21 corrected
## Array 22 corrected
## Array 23 corrected
## Array 24 corrected
## Array 25 corrected
## Array 26 corrected
## Array 27 corrected
## Array 28 corrected
## Array 29 corrected
## Array 30 corrected
## Array 31 corrected
## Array 32 corrected
y <- normalizeBetweenArrays(y, method="quantile")
class(y)
## [1] "EList"
## attr(,"package")
## [1] "limma"
3. 基因过滤
- 去除对照探针
- 去除匹配不到genesymbol的探针
- 去除不表达的探针，去除的标准是：至少在一半样本中高于背景，通过y(other)gIsWellAboveBG来判断。
- 我自己加上了一个，测到多次的基因，只保留一个探针。
Control <- y$genes$ControlType==1L;table(Control)
## Control
## FALSE TRUE
## 43529 1486
NoSymbol <- is.na(y$genes$GeneName);table(NoSymbol)
## NoSymbol
## FALSE
## 45015
IsExpr <- rowSums(y$other$gIsWellAboveBG > 0) >= 16;table(IsExpr)
## IsExpr
## FALSE TRUE
## 13088 31927
Isdup <- duplicated(y$genes$GeneName);table(Isdup)
## Isdup
## FALSE TRUE
## 30328 14687
yfilt <- y[!Control & !NoSymbol & IsExpr & !Isdup, ]
dim(yfilt)
## [1] 20650 32
可以看到，过滤后剩下了2万多个探针。
4.得到表达矩阵
exp = [email protected][[1]]
boxplot(exp)
image.png
exp[1:2,1:2]
## rawdata/GSE23558_RAW//GSM577914 rawdata/GSE23558_RAW//GSM577915
## [1,] 9.284154 11.473334
## [2,] 7.341236 7.474406
得到的表达矩阵没问题，但行名和列名均有问题。行名应该是探针名，列名是样本名，调整一下。
4.1获得样本名
colnames(exp) = str_extract(colnames(exp),"GSM\\d*")
exp[1:2,1:2]
## GSM577914 GSM577915
## [1,] 9.284154 11.473334
## [2,] 7.341236 7.474406
4.2 获得基因名
limma文档里写的是用了注释R包，在本例的原文件是里有探针注释的，这里直接使用。
可以直接将exp的行名转为基因名。行名不能重复，所以先去重
anno = yfilt$genes
nrow(anno);nrow(exp)
## [1] 20650
## [1] 20650
rownames(exp)=rownames(anno)
ids = unique(anno$GeneName)
exp = exp[!duplicated(anno$GeneName),]
rownames(exp) = ids
exp[1:4,1:4]
## GSM577914 GSM577915 GSM577916 GSM577917
## APOBEC3B 9.284154 11.473334 10.439071 11.661000
## A_32_P77178 7.341236 7.474406 7.310818 7.397149
## ATP11B 9.963452 8.915621 10.193873 9.321954
## DNAJA1 13.469790 13.201078 12.827357 13.389431
至此，得到了标准的表达矩阵。后面要做什么就看你啦，这就相当于修复了一下数据库里那个被标准化过的表达矩阵。
5.差异分析
design <- model.matrix(~group_list)
fit <- lmFit(exp,design)
fit <- eBayes(fit,trend=TRUE,robust=TRUE)
summary(decideTests(fit))
## (Intercept) group_listtumor
## Down 0 2102
## NotSig 0 16928
## Up 20650 1620
deg = topTable(fit,coef=2,n=dim(y)[1])
boxplot(exp[rownames(deg)[1],]~group_list)
image.png
save(exp,group_list,deg,gse,file = paste0(gse,"deg.Rdata"))