1. Next – generation Transcriptome Analysis Workshop
Lecture 4.1

2. Differential gene expression analysis
DESeq2 and edgeR
Reference
Transcriptome
De novo
Transcriptome
HTSeq
???
R Statistical Analysis Package
Differentially
Expressed
Genes
MA plots
Volcano plots
DESeq2
EdgeR
Counts

3. Differential gene expression analysis
The R Project
What is R???
How to run R
Unix command line
Rstudio
plots are a little more convenient
R console OK
slightly less convenient for plots and help
On PC, run as administrator
makes installing packages easier

4. Differential gene expression analysis
Bioconductor
load Bioconductor first, then use biocLite to load other packages
DESeq2
edgeR
> source("
Bioconductor version 3.1 (BiocInstaller ), ?biocLite for help
> biocLite("DESeq2")
BioC_mirror:
Using Bioconductor version 3.1 (BiocInstaller ), R version
Installing package(s) ‘DESeq2’
trying URL '
Content type 'application/zip' length bytes (3.2 MB)
downloaded 3.2 MB
package ‘DESeq2’ successfully unpacked and MD5 sums checked
The downloaded binary packages are in
C:\Users\Gribskov Admin\AppData\Local\Temp\RtmpeErDxa\downloaded_packages
Old packages: 'MASS', 'S4Vectors'
Update all/some/none? [a/s/n]: a
There are binary versions available but the source versions are later:
binary source needs_compilation
MASS TRUE
S4Vectors TRUE
Binaries will be installed
trying URL '
Content type 'application/zip' length bytes (1.0 MB)
downloaded 1.0 MB
trying URL '
Content type 'application/zip' length bytes (1.6 MB)
downloaded 1.6 MB
package ‘MASS’ successfully unpacked and MD5 sums checked
package ‘S4Vectors’ successfully unpacked and MD5 sums checked

5. Differential gene expression analysis
DESeq2
source("
biocLite("DESeq2")
library("DESeq2")
# set a working directory – this is on my desktop PC
directory <- "c:/users/mgribsko/Desktop/htseq"
setwd( directory)
# make a list of the count files produced by htseq, my files from htseq are called SRR<something>.count
files <- grep("SRR.*count",list.files(directory),value=TRUE)
# list a variable by typing its name
files
[1] "SRR count" "SRR count" "SRR count" "SRR count" "SRR count" "SRR count"
[7] "SRR count" "SRR count" "SRR count"
#alternatively, just list the file names by hand
files <- c("SRR count","SRR count","SRR count","SRR count","SRR count","SRR count",
+"SRR count","SRR count","SRR count")
# make a list of the sample names by extracting the substring before .count
samples <- sub("(*).count","\\1",files)
samples
[1] "SRR " "SRR " "SRR " "SRR " "SRR " "SRR " "SRR " "SRR " "SRR "
# create a dataframe, called matrix, with the sample names, files, and conditions
matrix <- data.frame(sampleName=samples, fileName=files, condition=c("wt","wt","wt","gpf1","gpf1","gpf1","cnf2","cnf2","cnf2"))
matrix
sampleName fileName condition
1 SRR SRR count wt
2 SRR SRR count wt
3 SRR SRR count wt
4 SRR SRR count gpf1
5 SRR SRR count gpf1
6 SRR SRR count gpf1
7 SRR SRR count cnf2
8 SRR SRR count cnf2
9 SRR SRR count cnf2
# read data into DESeq2, this is a special method designed to work with HTSeq files
ds<-DESeqDataSetFromHTSeqCount(sampleTable=matrix, directory=directory, design= ~ condition)

6. Differential gene expression analysis
DESeq2
some rows(genes) have no counts!
# reorder the "levels" in the data so wt is first. levels are the sample types
colData(ds)$condition<-factor(colData(ds)$condition,levels=c("wt","gpf1","cnf2")) ds
class: DESeqDataSet
dim:
exptData(0):
assays(1): counts
rownames(12709): XLOC_ XLOC_ XLOC_ XLOC_012709
rowData metadata column names(0):
colnames(9): SRR SRR SRR SRR
colData names(1): condition
#look at the first 10 rows
counts(ds[1:10,])
SRR SRR SRR SRR SRR SRR SRR SRR SRR
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
# total counts in each sample
colSums(counts(ds[,1:9]))
plot(counts(ds[,1]),counts(ds[,2]),pch=20)
plot(log10(counts(ds[,1])),log10(counts(ds[,2])),pch=20)
12709 rows, 9 columns

7. Differential gene expression analysis
DESeq2
# DESeq2 is run in a single command
ds <- DESeq(ds)
estimating size factors
estimating dispersions
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates
fitting model and testing
#results are extracted using the results function
res_gpf1_wt <- results(ds,contrast=c("condition","wt","gpf1"))
res_gpf1_wt
log2 fold change (MAP): condition wt vs gpf1
Wald test p-value: condition wt vs gpf1
DataFrame with rows and 6 columns
baseMean log2FoldChange lfcSE stat pvalue padj
<numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
XLOC_ NA
XLOC_ NA NA NA NA NA
XLOC_ NA
XLOC_
XLOC_ NA
XLOC_ NA NA NA NA NA
XLOC_ NA NA NA NA NA
XLOC_ NA NA NA NA NA
XLOC_
XLOC_ NA

8. Differential gene expression analysis
DESeq2
should be approximately log-normal, we see a big tail on the left (low expression) side
DESeq incorporates a test for outliers (Cook's cutoff)
# show in order of adjusted P Value (FDR)
res_gpf1_wt[order(res_gpf1_wt$padj),]
log2 fold change (MAP): condition wt vs gpf1
Wald test p-value: condition wt vs gpf1
DataFrame with rows and 6 columns
baseMean log2FoldChange lfcSE stat pvalue padj
<numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
XLOC_ e e+00
XLOC_ e e+00
XLOC_ e e-296
XLOC_ e e-280
XLOC_ e e-272
XLOC_ NA NA NA NA NA
XLOC_ NA NA NA NA NA
XLOC_ NA NA NA NA NA
XLOC_ NA NA NA NA NA
XLOC_ NA
# histogram of mean expression levels
hist(log(res_gpf1_wt[,"baseMean"]),breaks=seq(-5,15,by=0.25),main="gpf1 vs wt, no cutoff")

9. Differential gene expression analysis
DESeq2
# Apply Cook's cutoff
cut_gpf1_wt <- results(ds,contrast=c("condition","wt","gpf1"),cooksCutoff=1)
par(mfrow=c(2,1))
hist(log(cut_gpf1_wt[!is.na(cut_gpf1_wt[,"padj"]),"baseMean"]),breaks=seq(-5,15,by=0.25),main="gpf1 vs wt,cutoff=1")
hist(log(res_gpf1_wt[,"baseMean"]),breaks=seq(-5,15,by=0.25),main="gpf1 vs wt,no cutoff")

10. Differential gene expression analysis
DESeq2
# remove rows with adjusted P Value = NA (Cook's cutoff)
cut_gpf1_wt <- cut_gpf1_wt[!is.na(cut_gpf1_wt[,"padj"]),]
cut_gpf1_wt
log2 fold change (MAP): condition wt vs gpf1
Wald test p-value: condition wt vs gpf1
DataFrame with 8519 rows and 6 columns
baseMean log2FoldChange lfcSE stat pvalue padj
<numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
XLOC_ e e-01
XLOC_ e e-01
XLOC_ e e-02
XLOC_ e e-01
XLOC_ e e-14
XLOC_ e e-01
XLOC_ e e-07
XLOC_ e e-01
XLOC_ e e-01
XLOC_ e e-01
# Simple PCA to check whether the dat make sense, i.e., do the replicates seem more like each other than the different samples
vsd=varianceStabilizingTransformation(cut_gpf1_wt)
plotPCA(vsd,intgroup=c("condition"))
12709 rows -> 8519

11. Differential gene expression analysis
DESeq2
# MA plot is a traditional way to look at DGE results
plotMA(cut_gpf1_wt,ylim=c(-8,8))
abline(h=2,col="blue")
abline(h=-2,col="blue")
# volcano plots are another traditional method
with(cut_gpf1_wt,plot(log2FoldChange,-log10(padj),pch=20))
with(subset(cut_gpf1_wt,padj<0.01),points(log2FoldChange,-log10(padj),pch=20,col="red"))
with(subset(cut_gpf1_wt,abs(log2FoldChange)>3 & padj<0.01),points(log2FoldChange,-log10(padj),pch=20,col="green"))

12. Differential gene expression analysis
DESeq2
# plot estimated dispersion
plotDispEsts(ds)
# select significant genes with large fold change
select=cut_gpf1_wt[cut_gpf1_wt[,"padj"]<0.01 & abs(cut_gpf1_wt[,"log2FoldChange"])>4,]
select=select[order(select$log2FoldChange),]
# write results to file
write.table( select,file="c:/users/mgribsko/Desktop/htseq/edger.selected",row.names=T,col.names=T,sep="\t")

13. Differential gene expression analysis
edgeR
>ds # starting with the counts from htseq that were read into DESeq2
class: DESeqDataSet
dim:
exptData(0):
assays(1): counts
rownames(12709): XLOC_ XLOC_ XLOC_ XLOC_012709
rowRanges metadata column names(0):
colnames(9): SRR SRR SRR SRR
colData names(1): condition
mm = counts(ds) # extract the count matrix
head(mm)
SRR SRR SRR SRR SRR SRR SRR SRR SRR
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
> colSums( mm ) # Library Sizes
# some simple exploratory plots
hist(log(mm[,1]),breaks=100)

14. Differential gene expression analysis
edgeR
plot(log(mm[,1]),log(mm[,2]),pch=20, col=rgb(1,0,0,0.01))
pairs(log(mm[,1:9]),pch=20,cex=0.5)

15. Differential gene expression analysis
edgeR
# make the DGEList (data structure used by edgeR)
group=c("wt","wt","wt","gpf1","gpf1","gpf1","cnf2","cnf2","cnf2")
group
[1] "wt" "wt" "wt" "gpf1" "gpf1" "gpf1" "cnf2" "cnf2" "cnf2"
cds = DGEList( mm, group=group )
cds
An object of class "DGEList"
$counts
SRR SRR SRR SRR SRR SRR SRR SRR SRR
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
12704 more rows ...
$samples
group lib.size norm.factors
SRR wt
SRR wt
SRR wt
SRR gpf
SRR gpf
SRR gpf
SRR cnf
SRR cnf
SRR cnf
# filter counts that are too low to be reliable, the edgeR author's suggest something like the following
# cds <- cds[rowSums(1e+06 * cds$counts/expandAsMatrix(cds$samples$lib.size, dim(cds)) > 1) >= 3, ]
# I'll do something a little more complex, I want to keep any gene for which one or more samples have at least
# two values with at least 30 counts
cds=cds[(rowSums(mm[,1:3]>=30)>2|rowSums(mm[,4:6]>=30)>2|rowSums(mm[,7:9]>=30)>2),]
dim(cds)
[1] # 7620 of pass this filter

16. Differential gene expression analysis
edgeR
# get library normalization factors
cds <- calcNormFactors( cds )
cds$samples
group lib.size norm.factors
SRR wt
SRR wt
SRR wt
SRR gpf
SRR gpf
SRR gpf
SRR cnf
SRR cnf
SRR cnf
# do my replicates seem closer to each other than the samples
plotMDS( cds , main = "MDS Plot for Count Data", pch=20,cex=5,col=c(rep("red",3),rep("green",3),rep("blue",3)),top=1000 )
# estimate the dispersions
cds <- estimateCommonDisp(cds)
cds <- estimateTagwiseDisp( cds )

17. Differential gene expression analysis
edgeR
An object of class "DGEList"
$counts
SRR SRR SRR SRR SRR SRR SRR SRR SRR
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
$samples
group lib.size norm.factors
SRR wt
SRR wt
SRR wt
SRR gpf
SRR gpf
SRR gpf
SRR cnf
SRR cnf
SRR cnf
$common.dispersion
[1]
$pseudo.counts
XLOC_
XLOC_
XLOC_
XLOC_
XLOC_
$pseudo.lib.size
[1]
$AveLogCPM
[1]
7615 more elements ...
$prior.n
[1]
$tagwise.dispersion
[1]

18. Differential gene expression analysis
edgeR - dispersion
meanVarPlot <- plotMeanVar( cds , show.raw.vars=T,show.tagwise.vars=T,pch=19,show.binned.common.disp.vars=F,show.ave.raw.vars=F,
+ nbins = 100 ,xlab ="Mean Expression (Log10 Scale)" ,ylab = "Variance (Log10 Scale)" ,main = "Mean-Variance Plot")

19. Differential gene expression analysis
edgeR
# the significance test
gpf_wt=exactTest(cds,pair=c("wt","gpf1"))
# select the top DEGs
sel=abs(res[,"logFC"])>5 & res[,"PValue"]<0.01
selres=res[sel,]
selres = selres[order(selres[,"logFC"]),]
head(selres)
logFC logCPM PValue
XLOC_ e-233
XLOC_ e-28
XLOC_ e-26
XLOC_ e-71
XLOC_ e-45
XLOC_ e-21
# save the top DEGs
write.table( selres,file="c:/users/mgribsko/Desktop/edger.selected",row.names=T,col.names=T,sep="\t")

20. Differential gene expression analysis
edgeR - volcano plot
head(res)
logFC logCPM PValue
XLOC_ e+00
XLOC_ e+00
XLOC_ e-245
XLOC_ e-233
XLOC_ e-221
XLOC_ e-202
with(res,plot(logFC,-log10(PValue),pch=20))
with(subset(res,PValue<0.01),points(logFC,-log10(PValue),pch=20,col="red"))
with(subset(res,abs(logFC)>5),points(logFC,-log10(PValue),pch=20,col="orange"))
with(subset(res,abs(logFC)>5 & PValue<0.01),points(logFC,-log10(PValue),pch=20,col="green"))

21. Differential gene expression analysis
edgeR - MA plot (edgeR)
plot(res[,"logCPM"],res[,"logFC"],pch=20)
points(selres[,"logCPM"],selres[,"logFC"],pch=20,col="red")

22. Differential gene expression analysis
DESeq2 vs edgeR

23. Trinity - De novo transcriptome assembly
DEG analysis of Trinity results
Use RSEM to get counts
run RSEM on individual samples
gather in to counts matrix
# prepare reference
# rsem-prepare-reference [options] reference_fasta_file(s) reference_name
$TRINITY/trinity-plugins/rsem /rsem-prepare-reference \
--transcript-to-gene-map gene_map \
--bowtie2 \
../trinity_out_dir/Trinity.fasta \
magnaporthe.ref
# rsem-calculate-expression [options] upstream_read_file(s) reference_name sample_name
$TRINITY/trinity-plugins/rsem /rsem-calculate-expression -p 20 \
--bowtie2 --bowtie2-sensitivity-level very_sensitive \
../../cleaned/SRR trimmomatic.fq \
magnaporthe.ref \
magnaporte.wt.0.rsem
abundance_estimates_to_matrix.pl --est_method RSEM --out magnaporthe.trinity.rsem.gene \
magnaporte.wt.0.rsem.genes.results \
magnaporte.wt.1.rsem.genes.results \
magnaporte.wt.2.rsem.genes.results \
magnaporte.gpf1.0.rsem.genes.results \
magnaporte.gpf1.1.rsem.genes.results \
magnaporte.gpf1.2.rsem.genes.results \
magnaporte.cnf2.0.rsem.genes.results \
magnaporte.cnf2.1.rsem.genes.results \
magnaporte.cnf2.2.rsem.genes.results

24. Trinity - De novo transcriptome assembly
biocLite("DESeq2")
library("DESeq2")
directory <- "c:/users/mgribsko/Desktop/trinity"
setwd( directory)
t <- read.delim("magnaporthe.trinity.rsem.gene.counts.matrix",header=T,sep="\t",row.names=1)
head(t)
magnaporte.wt.0.rsem magnaporte.wt.1.rsem magnaporte.wt.2.rsem magnaporte.gpf1.0.rsem magnaporte.gpf1.1.rsem
TR8197|c0_all
TR4360|c0_all
TR16310|c0_all
TR4052|c0_all
TR9094|c0_all
TR21273|c0_all
magnaporte.gpf1.2.rsem magnaporte.cnf2.0.rsem magnaporte.cnf2.1.rsem magnaporte.cnf2.2.rsem
TR8197|c0_all
TR4360|c0_all
TR16310|c0_all
TR4052|c0_all
TR9094|c0_all
TR21273|c0_all
# rename the columns to be a little more convenient
colnames(t) <- sub("magnaporte\\.(.*)\\.rsem","\\1",colnames(t))
# set up sample metadata matrix
matrix=data.frame(sampleName=colnames(t),condition=c("wt","wt","wt","gpf1","gpf1","gpf1","cnf2","cnf2","cnf2"))
# read into DESeq data set
count=DESeqDataSetFromMatrix(round(t),colData=matrix,design=~condition,tidy=F)

25. Trinity - De novo transcriptome assembly
vsd=varianceStabilizingTransformation(count)
-- note: fitType='parametric', but the dispersion trend was not well captured by the
function: y = a/x + b, and a local regression fit was automatically substituted.
specify fitType='local' or 'mean' to avoid this message next time.
plotPCA(vsd,intgroup=c("condition))

26. Trinity - De novo transcriptome assembly
hist(log10(t[,1]),breaks=seq(-5,15,by=0.05))
pairs(log(t[,1:9]),pch=20,cex=0.5)
plot(log(t[,7]),log(t[,6]),pch=20,cex=0.5)

27. Trinity - De novo transcriptome assembly
count <- DESeq(count)
estimating size factors
estimating dispersions
gene-wise dispersion estimates
mean-dispersion relationship
final dispersion estimates
fitting model and testing
gpf1_wt=results(count,contrast=c("condition","gpf1","wt"),cooksCutoff=0.5)
table(is.na(gpf1_wt[,"padj"]))
FALSE TRUE
hist(log(gpf1_wt[!is.na(gpf1_wt[,"padj"]),"baseMean"]),breaks=seq(-5,15,by=0.25),main="gpf1 vs wt,cutoff=0.5")

28. Trinity - De novo transcriptome assembly
cut_gpf1_wt <- gpf1_wt[!is.na(gpf1_wt[,"padj"]),]
cut_gpf1_wt
log2 fold change (MAP): condition gpf1 vs wt
Wald test p-value: condition gpf1 vs wt
DataFrame with rows and 6 columns
baseMean log2FoldChange lfcSE stat pvalue padj
<numeric> <numeric> <numeric> <numeric> <numeric> <numeric>
TR16310|c0_all e e-01
TR4052|c0_all e e-02
TR9094|c0_all e e-01
TR21273|c0_all e e-19
TR2841|c0_all e e-66
TR10766|c0_all
TR12328|c0_all
TR18453|c0_all
TR14941|c0_all
TR11481|c0_all
plotMA(cut_gpf1_wt,ylim=c(-8,8))
abline(h=2,col="blue")
abline(h=-2,col="blue")
# select DEGs
select=cut_gpf1_wt[cut_gpf1_wt[,"padj"]<0.01 & abs(cut_gpf1_wt[,"log2FoldChange"])>4,]
select=select[order(select$log2FoldChange),]
# write results to file
write.table( select,file="c:/users/mgribsko/Desktop/trinity/trinity_deseq.selected",row.names=T,col.names=T,sep="\t")

29. Trinity - De novo transcriptome assembly
reference

30. Trinity - De novo transcriptome assembly
edger_gpf1_wt=exactTest(cds,pair=c("wt","gpf1"))
edger_gpf1_wt
An object of class "DGEExact"
$table
logFC logCPM PValue
TR16310|c0_all e-01
TR4052|c0_all e-02
TR21273|c0_all e-22
TR2841|c0_all e-37
TR6873|c0_all e-01
10114 more rows ...
$comparison
[1] "wt" "gpf1"
$genes
NULL
sel = edger_gpf1_wt$table
dim(sel)
[1]
sel=sel[abs(sel[,"logFC"])>5 & sel[,"PValue"]<0.01,]
[1]
sel=sel[order(sel[,"logFC"]),]
head(sel) logFC logCPM PValue
TR21909|c0_all e-181
TR24805|c0_all e-18
TR14368|c0_all e-16
TR6143|c0_all e-45
TR20597|c0_all e-04
TR20018|c0_all e-13
t["TR24805|c0_all",]
wt.0 wt.1 wt.2 gpf1.0 gpf1.1 gpf1.2 cnf2.0 cnf2.1 cnf2.2
TR24805|c0_all
write.table(sel,file="c:/users/mgribsko/Desktop/trinity_edger.selected",row.names=T,col.names=T,sep="\t")