学术咖

TCseq可根据不同的聚类方法将基因按照表达模式分类

BiocManager::install("TCseq")
library(TCseq)
data <- read.delim('df_TCseq.xls', row.names = 1, sep = '\t', check.names = FALSE)
data <- as.matrix(data)
tca <- timeclust(data, algo = "cm", k = 6, standardize = TRUE)
#character string giving a clustering method. Options are km' (kmeans), 'pam' (partitioning around medoids), 'hc' (hierachical clustering), 'cm' (cmeans).
p <- timeclustplot(tca, value = "z-score(TPM)", cols = 3)#所有cluster合并一个图
print(p[[1]])#单一cluster作图
a <- as.data.frame(tca@cluster)
table(a)
names(a) <- 'Cluster'
Cluster2 <- subset(a,Cluster == 2)
write.table(Cluster2, file="Cluster2_gene.xls", sep="\t", quote=F, row.names=T, col.names=T)

导出不同cluster基因做功能富集

DNA甲基化：DNA甲基化为DNA化学修饰的一种形式，能在不改变DNA序列的前提下，改变遗传表现。为表观遗传编码的一部分，是一种外遗传机制。DNA甲基化过程会使甲基添加到DNA分子上，例如在胞嘧啶环的5’碳上：这种5’方向的DNA甲基化方式可见于所有脊椎动物。 在人类细胞内，大约有1%的DNA碱基受到了甲基化。

gzip -dc A.1_bismark_bt2_pe.deduplicated.CX_report.txt.gz > A.1_bismark_bt2_pe.deduplicated.CX_report.txt
perl BismarkCX2methykit.pl A.1_bismark_bt2_pe.deduplicated.CX_report.txt
#准备注释需要的bed文件（格式12）
convert2bed -i gtf < out.gtf > out6.bed
grep 'exon' out6.bed > aa && mv aa out6.bed
python3 bed6Tobed12.py out6.bed  > out12.bed#bed6Tobed12.py
#methylKit安装及使用
if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("methylKit")
library(methylKit)
file.list = list("A.CG_methykit.txt","B.CG_methykit.txt","C.CG_methykit.txt","D.CG_methykit.txt","E.CG_methykit.txt","F.CG_methykit.txt","G.CG_methykit.txt","H.CG_methykit.txt","I.CG_methykit.txt")
myobjDB=methRead(file.list,
sample.id=list("Control_1","Control_2","Control_3","Single_1","Single_2","Single_3","Multiple_1","Multiple_2","Multiple_3"),
assembly="HZGC",
treatment=c(0,0,0,1,1,1,2,2,2),
context="CpG",
mincov = 10,
dbtype = "tabix",
dbdir = "methylDB",
)
filtered.myobj=filterByCoverage(myobjDB,lo.count=10,lo.perc=NULL, hi.count=NULL,hi.perc=99.9)
meth=unite(filtered.myobj, destrand=FALSE)

pdf("Correlation.pdf")
getCorrelation(meth,plot=TRUE)
dev.off()
pdf("clusterSamples.pdf")
clusterSamples(meth, dist="correlation", method="ward", plot=TRUE)
dev.off()
pdf("PCAscreeplot.pdf")
PCASamples(meth, screeplot=TRUE)
dev.off()
pdf("PCAcluster.pdf")
PCASamples(meth)
dev.off()

tiles=tileMethylCounts(filtered.myobj,win.size=1000,step.size=1000)
meth=unite(tiles, destrand=FALSE)
myDiff=calculateDiffMeth(meth,num.cores=10)
diffMethPerChr(myDiff,plot=FALSE,qvalue.cutoff=0.01, meth.cutoff=25)
myDiff25p.hyper=getMethylDiff(myDiff,difference=25,qvalue=0.01,type="hyper")
myDiff25p.hypo=getMethylDiff(myDiff,difference=25,qvalue=0.01,type="hypo")
myDiff25p=getMethylDiff(myDiff,difference=25,qvalue=0.01)
diffMethPerChr(myDiff,plot=FALSE,qvalue.cutoff=0.01, meth.cutoff=25)
save.image("MS_M_S.RData")
Diffdataframe=getData(myDiff25p)
write.table(Diffdataframe, file="Diffdataframe.xls", sep="\t", quote=F, row.names=T, col.names=T)

data <- read.delim('mouse_gc_log2.xls', row.names = 1, sep = '\t', check.names = FALSE)#行为基因ID列为样本的表达矩阵
data[order(rowSums(data),decreasing=T)[1:3000],]  #筛选表达量高的前3000个基因
data[rowSums(data)>1,]  #过滤掉表达量低的基因

data1=rowMeans(data)>1  #按平均数筛选
data2=rowSums(data>0)>6 #表达量不为0的样品个数筛选
data=data[data1 & data2,] #联合一下
data$cv <- apply(data, 1, function(x){
  sd(x)/mean(x)*100
})
data_df <- data[order(data$cv, decreasing = T)[1:3000], 1:15]#筛选变异系数最大的3000个基因

library(edgeR)
library(ggplot2)
data <- read.delim('../mouse_TPM_notCross.matrix.xls', row.names = 1, sep = '\t', check.names = FALSE)
group <- factor(rep(c('A', 'B', 'C'), each = 3))
y <- DGEList(counts = data, group = group)
y <- calcNormFactors(y)
logcpm <- cpm(y, prior.count=3, log=TRUE)
write.table(logcpm, file="mouse_TMM.xls", sep="\t", quote=F, row.names=T, col.names=T)
#dgList <- estimateCommonDisp(y)
#dgList <- estimateTagwiseDisp(dgList)
#norm_counts.table <- t(t(dgList$pseudo.counts)*(dgList$samples$norm.factors))
#write.table(norm_counts.table, file="mouse_gc_pareto_TMM.xls", sep="\t", quote=F)
#pheatmap(log(data+1),cluster_rows=T,cluster_cols=T,scale="none",border_color="white",color=colorRampPalette(rev(c("red","white","blue")))(102))

什么是批次效应（batch effect）？
不同平台的数据，同一平台的不同时期的数据，同一个样品不同试剂的数据，以及同一个样品不同时间的数据等等都会产生一种batch effect 。这种影响如果广泛存在应该被足够重视，否则会导致整个实验和最终的结论失败。比对实验组和对照组，不同的处理是患病和不患病（测序时，先测得疾病，然后测得正常），然后你通过分析，得到很多差异表达的基因。现在问题来了，这个差异表达的结果是和你要研究的因素有关，还是时间有关，这个问题里时间就会成为干扰实验结果的因素，这个效应就是batch effect。

library(limma)
data <- read.table("mouse_gc.xls",sep="\t",header = TRUE)
batch1 <- c(rep('fish',9),rep('mouse',9))
batch1 <- as.factor(batch1)
design <- model.matrix(~0 + batch1)
#data=normalizeBetweenArrays(data)#如果组内的中位数不在同一条水平线上，现在该参数校正，再使用下面的批次校正
new_data <- removeBatchEffect(data[,2:19], batch = batch1)
boxplot(data[,2:19])
boxplot(new_data)
write.table(new_data, file="removeBatch.xls", sep="\t", quote=F, row.names=T, col.names=T)

limma结果出现负值请参考：https://cloud.tencent.com/developer/article/1680305
参考链接：https://www.omicsclass.com/article/1113

library("preprocessCore")
library("ggplot2")
data <- read.table("mouse_gc.xls",sep="\t",header = TRUE)
rownames(data) <- data$GeneID
#rownames(data) <- data[,1]
#data_mat <- data.matrix(data[,-1]) 
quantile_normalisation <- function(df){
  df_rank <- apply(df,2,rank,ties.method="min")
  df_sorted <- data.frame(apply(df, 2, sort))
  df_mean <- apply(df_sorted, 1, mean)

  index_to_mean <- function(my_index, my_mean){
    return(my_mean[my_index])
  }

  df_final <- apply(df_rank, 2, index_to_mean, my_mean=df_mean)
  rownames(df_final) <- rownames(df)
  return(df_final)
}
new_data <- quantile_normalisation(data[,2:19])
boxplot(data)
boxplot(new_data)
write.table(new_data, file="111.xls", sep="\t", quote=F, row.names=T, col.names=T)
##############################################################
col_names <- colnames(data)
row_names <- rownames(data)
data <- as.matrix(data)
b=normalize.quantiles(data)
boxplot(b)
##使用normalize.quantiles函数和上面结果相同