Testing for differential expression 2014

Objectives

Once we've obtained abundance counts for our genes/exons/transcripts, we are usually interested in identifying those genes/exons/transcripts that are differentially expressed. 

In this section, you will learn about 4 different tools for identifying differentially expressed genes from gene count data. The same data from the previous exercises will be used. The data contains 75 bp paired-end reads that have been generated in silico to replicate real gene count data from Drosophila. The data simulates two biological groups with three biological replicates per group (6 samples total). 

• Learn about DESeq, edgeR, DEXSeq and cuffdiff packages and the differences among these packages.
• Become familiar with basic R usage and installing Bioconductor modules.
• Learn how to use edgeR/DESeq to identify differentially expressed genes.
• Learn how to use cuffdiff pacakge to identify differentially expressed genes.

Get set up

cds

cd my_rnaseq_course
cd gene_expression_exercise 
# you should have already copied this over
 
#modify our gene counts file a bit
grep '^F' gene_counts.htseq.gff > gene_counts.htseq.final.gff

Let's first start installing some R tools because they may take a while. 

Introduction

Most RNA-Seq experiments are conducted with the aim of identifying genes/exons that are differentially expressed between two or more conditions. Many computational tools are available for performing the statistical tests required to identify these genes/exons.

Simply put, all these tools do three steps (and they can vary in how they do these steps):

• Normalization of gene counts
  
  
• Represent the gene counts by a distribution that defines the relation between mean and variance (dispersion).
  
  
• Perform a statistical test on this distribution to identify genes that are significantly different between the conditions.
  
  
• Provide fold change,  P-value information, false discovery rate for each gene.

Why Normalize?

Normalization smooths out technical variations among the samples we are comparing so that we can more confidently attribute variations we see to biological reasons. 

We usually normalize for:

1. Sequencing depth: Say we are comparing gene counts in sample A against sample B.  If you start out with 10 million reads in sample A  vs 1 million reads in sample B, a 10 fold increase in expression in sample A is going to be purely due to its sequencing depth.
2. Gene length: A gene that is twice as long is likely to have twice as many reads sampling it.

Most commonly done normalization

RPKM: Normalizes for sequencing depth and gene length.

RPKM = reads per kilobase per million mapped reads

RPK= No.of Mapped reads/ length of transcript in kb (transcript length/1000)

RPKM = RPK/total no.of reads in million (total no of reads/ 1000000)

DEG tools compared

From paper: Wesolowski et al, Biosensors, 2013

R and Bioconductor, very briefly...

R is a very common scripting language used in statistics. There are whole courses on using R going on in other SSI classrooms as we speak! Inside the R universe, you have access to an incredibly large number of useful statistical functions (Fisher's exact test, nonlinear least-squares fitting, ANOVA ...). R also has advanced functionality for producing plots and graphs as output. 

Regrettably, R is a bit of it's own bizarro world, as far as how its commands work. (Futhermore, Googling "R" to get help can be very frustrating.) The conventions of most other programming and scripting languages seem to have been re-invented by someone who wanted to do everything their own way in R. Just like we write shell scripts in bash, you can write R scripts that carry out complicated analyses.

Hints for working with R

• Don't forget: it's q() to quit.
• For help with a function, type ?command. Try ?read.table. The q key gets you out of help, just like for a man page.
• The left arrow <- (less-than-dash) is the same as an equals sign =. You can use them interchangeably.
• The prompt we will sometimes be showing for R is >. Don't type this for a command. It is like the login1$ at the beginning of the bash prompt when you log in to Lonestar. It just means that you are in the R shell.

You can type the name of a variable to have its value displayed. Like this...

> x <- 10 + 5 + 6
> x
[1] 21

Bioconductor packages for R

Like other languages, R can be expanded by loading packages. The R equivalent of Bioperl or Biopython is Bioconductor. Bioconductor can theoretically do things for you like convert sequences (none of us use it for that), but where it really shines is in doing statistical tests (where is it second-to-none in this list of languages). Many functions for analyzing microarray data are implemented in R, and this strength has now carried over to the analysis of RNAseq data.

Here's how you install two modules that we will need for this exercise: (But first, remember that to get R on lonestar, you'll need to load the R module first)

login1$ module load R
login1$ R

> source("http://bioconductor.org/biocLite.R")
Warning in install.packages("BiocInstaller", repos = a["BioCsoft", "URL"]) :
  'lib = "/opt/apps/R/2.15.3/lib64/R/library"' is not writable
Would you like to use a personal library instead?  (y/n) y
Would you like to create a personal library
~/R/x86_64-unknown-linux-gnu-library/2.15
to install packages into?  (y/n) y
...
> biocLite("DESeq")
...
> biocLite("edgeR")
...
> q()
Save workspace image? [y/n/c]: n

When you start R later, you will not need to re-install the modules. You can load them with just these commands:

login1$ R
> library("DESeq")
> library("edgeR")

login1$ R
#LOAD LIBRARY, READ IN GENE COUNTS, ADD METADATA
> library("DESeq")
> counts = read.delim("gene_counts.htseq.final.gff", header=F, row.names=1)
> head(counts)
> colnames(counts) = c("C11","C12","C13","C21","C22","C23")
> head(counts)
> my.design <- data.frame(
  row.names = colnames( counts ),
  condition = c( "C1", "C1", "C1", "C2", "C2", "C2"),
  libType = c( "paired-end", "paired-end", "paired-end", "paired-end", "paired-end", "paired-end" )
)
> my.design
> conds <- factor(my.design$condition)
 
 
#CREATE A DESEQ DATA OBJECT FROM COUNTS
> cds <- newCountDataSet( counts, conds )
> cds

 
#NORMALIZATION: ESTIMATE SIZE FACTORS
> cds <- estimateSizeFactors( cds )
> sizeFactors( cds )
> head( counts( cds, normalized=TRUE ) )

 
#ESTIMATE DISPERSION/VARIANCE
> cds <- estimateDispersions( cds )


#DO TEST FOR DIFFERENTIAL EXPRESSION AND WRITE RESULTS INTO FILE
> result <- nbinomTest( cds, "C1", "C2" )
> head(result)
> result = result[order(result$pval), ]
> head(result)
> write.csv(result, "DESeq-C1-vs-C2.csv")


#GENERATE MA PLOT
> pdf("DESeq-MA-plot.pdf")
> plot(
  result$baseMean,
  result$log2FoldChange,
  log="x", pch=20, cex=.3,
  col = ifelse( result$padj < .1, "red", "black" ) )
> dev.off()
> q()

Save workspace image? [y/n/c]: n
login1$ head DESeq-wt-vs-mut.csv

login1$ R

#LOAD LIBRARY, READ IN COUNTS, ADD SOME METADATA
> library("edgeR")
> counts = read.delim("gene_counts.htseq.gff", header=F, row.names=1)
> colnames(counts) = c("C11", "C12", "C13", "C21", "C22", "C23")
> head(counts)
> group <- factor(c("C1", "C1", "C1", "C2", "C2", "C2"))
 
 
#CREATE EDGER SPECIFIC DATA OBJECT FROM COUNTS DATA
> dge = DGEList(counts=counts,group=group)
 
 
#ESTIMATE DISPERSIONS(VARIANCE)
> dge <- estimateCommonDisp(dge)
> dge <- estimateTagwiseDisp(dge)
 
 
#DO TEST FOR DIFFERENTIAL EXPRESSION
> et <- exactTest(dge)
> etp <- topTags(et, n=100000)

 
#GENERATE MA PLOT
> pdf("edgeR-MA-plot.pdf")
> plot(
  etp$table$logCPM,
  etp$table$logFC,
  xlim=c(-3, 20), ylim=c(-12, 12), pch=20, cex=.3,
  col = ifelse( etp$table$FDR < .1, "red", "black" ) )
> dev.off()

 
#WRITE OUT GENE RESULTS
> write.csv(etp$table, "edgeR-wt-vs-mut.csv")
> q()
Save workspace image? [y/n/c]: y
login1$ head edgeR-wt-vs-mut.csv

head DESeq-wt-vs-mut.csv
head edgeR-wt-vs-mut.csv

#DESeq results
sed 's/,/\t/g' DESeq-C1-vs-C2.csv|sort -n -r -k6,6|cut -f 2,6|head

#edgeR results
sed 's/,/\t/g' edgeR-wt-vs-mut.csv |sort -n -r -k2,2|cut -f 1,2|head

#We will use log 2 fold change instead of fold change because edgeR doesnt report fold change
#DESeq results
sed 's/,/\t/g' DESeq-C1-vs-C2.csv|awk '{if (($7>=1)&&($8<=0.05)) print $1,$7,$8}'|head
sed 's/,/\t/g' DESeq-C1-vs-C2.csv|awk '{if (($7>=1)&&($8<=0.05)) print $1,$7,$8}'|wc -l

#edgeR results
sed 's/,/\t/g' edgeR-wt-vs-mut.csv |awk '{if (($2>=1)&&($3<=0.05)) print $1,$2,$3}'|head
sed 's/,/\t/g' edgeR-wt-vs-mut.csv |awk '{if (($2>=1)&&($3<=0.05)) print $1,$2,$3}'|wc -l