---
title: Workflow Demo for Ribo-Seq and polyRibo-Seq 
author: "First/last name (first.last@ucr.edu)"
date: "Last update: `r format(Sys.time(), '%d %B, %Y')`" 
output:
  BiocStyle::html_document:
    toc: true
    toc_depth: 3
    fig_caption: yes

fontsize: 14pt
bibliography: bibtex.bib
---
<!--
%% \VignetteEngine{knitr::rmarkdown}
%\VignetteIndexEntry{RIBO-Seq Workflow Template}
%% \VignetteDepends{rjson, ggplot2, limma, edgeR, DESeq2, GOstats, GO.db, annotate, pheatmap}
%% \VignetteKeywords{compute cluster, pipeline, reports}
%% \VignettePackage{systemPipeR}
-->

<!---
- Compile from command-line
Rscript -e "rmarkdown::render('systemPipeRIBOseq.Rmd')"; R CMD Stangle systemPipeRIBOseq.Rmd
-->

```{r style, echo = FALSE, results = 'asis'}
BiocStyle::markdown()
options(width=100, max.print=1000)
knitr::opts_chunk$set(
    eval=as.logical(Sys.getenv("KNITR_EVAL", "TRUE")),
    cache=as.logical(Sys.getenv("KNITR_CACHE", "TRUE")),
    warning=FALSE, message=FALSE)
```

Note: the most recent version of this tutorial can be found <a href="https://htmlpreview.github.io/?https://raw.githubusercontent.com/tgirke/systemPipeRdata/master/inst/extdata/workflows/riboseq/systemPipeRIBOseq.html">here</a> and a short overview slide show [here](https://docs.google.com/presentation/d/175aup31LvnbIJUAvEEoSkpGsKgtBJ2RpQYd0Gs23dLo/embed?start=false&loop=false&delayms=60000).


# Introduction

Ribo-Seq and polyRibo-Seq are a specific form of RNA-Seq gene expression
experiments utilizing mRNA subpopulations directly bound to ribosomes.
Compared to standard RNA-Seq, their readout of gene expression provides a
better approximation of downstream protein abundance profiles due to their
close association with translational processes. The most important difference
among the two is that polyRibo-Seq utilizes polyribosomal RNA for sequencing,
whereas Ribo-Seq is a footprinting approach restricted to sequencing RNA
fragments protected by ribosomes [@Ingolia2009-cb; @Aspden2014-uu; @Juntawong2015-ru}. 

The workflow presented in this vignette contains most of the data analysis
steps described by [@Juntawong2014-ny] including functionalities useful for
processing both polyRibo-Seq and Ribo-Seq experiments. To improve re-usability
and adapt to recent changes of software versions (_e.g._ R, Bioconductor and
short read aligners), the code has been optimized accordingly. Thus, the
results obtained with the updated workflow are expected to be similar but not
necessarily identical with the published results described in the original
paper. 

Relevant analysis steps of this workflow include read preprocessing, read
alignments against a reference genome, counting of reads overlapping with a
wide range of genomic features (_e.g._ CDSs, UTRs, uORFs, rRNAs, etc.),
differential gene expression and differential ribosome binding analyses, as
well as a variety of genome-wide summary plots for visualizing RNA expression
trends. Functions are provided for evaluating the quality of Ribo-seq data,
for identifying novel expressed regions in the genomes, and for gaining
insights into gene regulation at the post-transcriptional and translational
levels. For example, the functions `genFeatures` and
`featuretypeCounts` can be used to quantify the expression output for
all feature types included in a genome annotation (`e.g.` genes,
introns, exons, miRNAs, intergenic regions, etc.). To determine the approximate
read length of ribosome footprints in Ribo-Seq experiments, these feature type
counts can be obtained and plotted for specific read lengths separately.
Typically, the most abundant read length obtained for translated features
corresponds to the approximate footprint length occupied by the ribosomes of a
given organism group. Based on the results from several Ribo-Seq studies, these
ribosome footprints are typically ~30 nucleotides long
[@Ingolia2011-fc; @Ingolia2009-cb; @Juntawong2014-ny].  However, their
length can vary by several nucleotides depending upon the optimization of the
RNA digestion step and various factors associated with translational
regulation.  For quality control purposes of Ribo-Seq experiments it is also
useful to monitor the abundance of reads mapping to rRNA genes due to the high
rRNA content of ribosomes. This information can be generated with the 
`featuretypeCounts` function described above.

Coverage trends along transcripts summarized for any number of transcripts can
be obtained and plotted with the functions `featureCoverage` and
`plotfeatureCoverage`, respectively. Their results allow monitoring
of the phasing of ribosome movements along triplets of coding sequences.
Commonly, high quality data will display here for the first nucleotide of each
codon the highest depth of coverage computed for the 5' ends of the aligned
reads. 
 
Ribo-seq data can also be used to evaluate various aspects of translational
control due to ribosome occupancy in upstream open reading frames (uORFs). The
latter are frequently present in (or near) 5' UTRs of transcripts. For this,
the function `predORFs` can be used to identify ORFs in the
nucleotide sequences of transcripts or their subcomponents such as UTR regions.
After scaling the resulting ORF coordinates back to the corresponding genome
locations using `scaleRanges`, one can use these novel features
(_e.g._ uORFs) for expression analysis routines similar to those
employed for pre-existing annotations, such as the exonic regions of genes. For
instance, in Ribo-Seq experiments one can use this approach to systematically identify all
transcripts occupied by ribosomes in their uORF regions. The binding of
ribosomes to uORF regions may indicate a regulatory role in the translation of
the downstream main ORFs and/or translation of the uORFs into functionally
relevant peptides. 

## Experimental design
Typically, users want to specify here all information relevant for the analysis
of their NGS study. This includes detailed descriptions of FASTQ files,
experimental design, reference genome, gene annotations, etc.  

# Load workflow template

The following loads one of the available NGS workflow templates (here RIBO-Seq)
into the user's current working directory. At the moment, the package includes
workflow templates for RNA-Seq, ChIP-Seq, VAR-Seq and Ribo-Seq. Templates for
additional NGS applications will be provided in the future.
The sample data are described [here](http://www.bioconductor.org/packages/devel/bioc/vignettes/systemPipeR/inst/doc/systemPipeR.html#load-sample-data-and-workflow-templates).

```{r genRibo_workflow, eval=FALSE}
library(systemPipeRdata)
genWorkenvir(workflow="riboseq", bam=TRUE)
setwd("riboseq")
```
## Download latest version of this tutorial

In case there is a newer version of this tutorial, download its `systemPipeRIBOseq.Rmd` source and open it in your R IDE (e.g. vim-r or RStudio).


```{r download_latest, eval=FALSE}
download.file("https://raw.githubusercontent.com/tgirke/systemPipeRdata/master/inst/extdata/workflows/riboseq/systemPipeRIBOseq.Rmd", "systemPipeRIBOseq.Rmd")
```

## Load packages and sample data
The `systemPipeR` package needs to be loaded to perform the analysis 
steps shown in this report [@Girke2014-oy]. The package allows users
to run the entire analysis workflow interactively or with a single command 
while also generating the corresponding analysis report. For details
see `systemPipeR's` main [vignette](http://www.bioconductor.org/packages/devel/bioc/vignettes/systemPipeR/inst/doc/systemPipeR.html).

```{r load_systempiper, eval=TRUE}
library(systemPipeR)
```

In the workflow environments generated by `genWorkenvir` all data
inputs are stored in a `data/` directory and all analysis results will
be written to a separate `results/` directory, while the
`systemPipeRIBOseq.Rmd` script and the `targets` file are
expected to be located in the parent directory. The R session is expected to
run from this parent directory. Additional parameter files are stored under
`param/`.

To work with real data, users want to organize their own data similarly and
substitute all test data for their own data. To rerun an established workflow
on new data, the initial `targets` file along with the corresponding
FASTQ files are usually the only inputs the user needs to provide.

If applicable users can load custom functions not provided by
`systemPipeR`. Skip this step if this is not the case.

```{r source_helper_fcts, eval=FALSE}
source("systemPipeRIBOseq_Fct.R")
```

## Experiment definition provided by `targets` file
The `targets` file defines all FASTQ files and sample comparisons of the analysis workflow. 

```{r load_targets, eval=TRUE}
targetspath <- system.file("extdata", "targets.txt", package="systemPipeR")
targets <- read.delim(targetspath, comment.char = "#")[,1:4]
targets
```

# Read preprocessing

## Quality filtering and adaptor trimming
The following custom function trims adaptors hierarchically from the longest to
the shortest match of the right end of the reads. If
`internalmatch=TRUE` then internal matches will trigger the same
behavior.  The argument `minpatternlength` defines the shortest
adaptor match to consider in this iterative process. In addition, the function
removes reads containing Ns or homopolymer regions. More detailed information
on read preprocessing is provided in `systemPipeR's` main vignette.

```{r fastq_trimming, eval=FALSE}
args <- systemArgs(sysma="param/trim.param", mytargets="targets.txt")
fctpath <- system.file("extdata", "custom_Fct.R", package="systemPipeR")
source(fctpath)
iterTrim <- ".iterTrimbatch1(fq, pattern='ACACGTCT', internalmatch=FALSE, minpatternlength=6, 
                             Nnumber=1, polyhomo=50, minreadlength=16, maxreadlength=101)" 
preprocessReads(args=args, Fct=iterTrim, batchsize=100000, overwrite=TRUE, compress=TRUE)
writeTargetsout(x=args, file="targets_trim.txt", overwrite=TRUE)
```

## FASTQ quality report
The following `seeFastq` and `seeFastqPlot` functions generate and plot a series of
useful quality statistics for a set of FASTQ files including per cycle quality
box plots, base proportions, base-level quality trends, relative k-mer
diversity, length and occurrence distribution of reads, number of reads above
quality cutoffs and mean quality distribution. The results are written to a PDF file named 
`fastqReport.png`.

```{r fastq_report, eval=FALSE}
args <- systemArgs(sysma=NULL, mytargets="targets_trim.txt")
fqlist <- seeFastq(fastq=infile1(args), batchsize=100000, klength=8)
png("./results/fastqReport.png", height=18, width=4*length(fqlist), units="in", res=72)
seeFastqPlot(fqlist)
dev.off()
```

![](results/fastqReport.png)
<div align="center">Figure 1: FASTQ quality report for 18 samples</div></br>

# Alignments

## Read mapping with `Bowtie2/Tophat2`
The NGS reads of this project will be aligned against the reference genome
sequence using `Bowtie2/TopHat2` [@Kim2013-vg; @Langmead2012-bs]. The
parameter settings of the aligner are defined in the `tophat.param`
file.

```{r tophat_alignment1, eval=FALSE}
args <- systemArgs(sysma="param/tophat.param", mytargets="targets_trim.txt")
sysargs(args)[1] # Command-line parameters for first FASTQ file
```

Submission of alignment jobs to compute cluster, here using 72 CPU cores (18 `qsub` processes each with 4 CPU cores).

```{r tophat_alignment2, eval=FALSE, warning=FALSE, message=FALSE}
moduleload(modules(args))
system("bowtie2-build ./data/tair10.fasta ./data/tair10.fasta")
resources <- list(walltime="20:00:00", nodes=paste0("1:ppn=", cores(args)), memory="10gb")
reg <- clusterRun(args, conffile=".BatchJobs.R", template="torque.tmpl", Njobs=18, runid="01", 
                  resourceList=resources)
waitForJobs(reg)
```

## Read mapping with `HISAT2`

```{r hisat2_alignment, eval=FALSE}
args <- systemArgs(sysma="param/hisat2.param", mytargets="targets.txt")
# args <- systemArgs(sysma="param/hisat2.param", mytargets="targets_trim.txt")
sysargs(args)[1] # Command-line parameters for first FASTQ file
moduleload(modules(args))
system("hisat2-build ./data/tair10.fasta ./data/tair10.fasta")
runCommandline(args=args)
```

Check whether all BAM files have been created

```{r check_files_exist, eval=FALSE}
file.exists(outpaths(args))
```

## Read and alignment stats
The following provides an overview of the number of reads in each sample and how many of them aligned to the reference.

```{r align_stats, eval=FALSE}
read_statsDF <- alignStats(args=args) 
write.table(read_statsDF, "results/alignStats.xls", row.names=FALSE, quote=FALSE, sep="\t")
```

```{r align_stats_view, eval=TRUE}
read.table(system.file("extdata", "alignStats.xls", package="systemPipeR"), header=TRUE)[1:4,]
```

## Create symbolic links for viewing BAM files in IGV

The `symLink2bam` function creates symbolic links to view the BAM alignment files in a
genome browser such as IGV. The corresponding URLs are written to a file
with a path specified under `urlfile` in the `results` directory.

```{r bam_urls, eval=FALSE}
symLink2bam(sysargs=args, htmldir=c("~/.html/", "projects/tests/"), 
            urlbase="http://biocluster.ucr.edu/~tgirke/", 
	        urlfile="./results/IGVurl.txt")
```

# Read distribution across genomic features

The `genFeatures` function generates a variety of feature types from
`TxDb` objects using utilities provided by the `GenomicFeatures` package. 

## Obtain feature types

The first step is the generation of the feature type ranges based on
annotations provided by a GFF file that can be transformed into a
`TxDb` object. This includes ranges for mRNAs, exons, introns, UTRs,
CDSs, miRNAs, rRNAs, tRNAs, promoter and intergenic regions. In addition, any
number of custom annotations can be included in this routine.

```{r genFeatures, eval=FALSE}
library(GenomicFeatures)
txdb <- makeTxDbFromGFF(file="data/tair10.gff", format="gff3", organism="Arabidopsis")
feat <- genFeatures(txdb, featuretype="all", reduce_ranges=TRUE, upstream=1000, downstream=0, 
                    verbose=TRUE)
```

## Count and plot reads of any length
The `featuretypeCounts` function counts how many reads in short read
alignment files (BAM format) overlap with entire annotation categories. This
utility is useful for analyzing the distribution of the read mappings across
feature types, _e.g._ coding versus non-coding genes. By default the
read counts are reported for the sense and antisense strand of each feature
type separately. To minimize memory consumption, the BAM files are processed in
a stream using utilities from the `Rsamtools` and
`GenomicAlignment` packages.  The counts can be reported for each read
length separately or as a single value for reads of any length.  Subsequently,
the counting results can be plotted with the associated
`plotfeaturetypeCounts` function.

The following generates and plots feature counts for any read length.

```{r featuretypeCounts, eval=FALSE}
library(ggplot2); library(grid)
fc <- featuretypeCounts(bfl=BamFileList(outpaths(args), yieldSize=50000), grl=feat, 
                        singleEnd=TRUE, readlength=NULL, type="data.frame")
p <- plotfeaturetypeCounts(x=fc, graphicsfile="results/featureCounts.png", graphicsformat="png", 
                           scales="fixed", anyreadlength=TRUE, scale_length_val=NULL)
```

![](results/featureCounts.png)
<div align="center">Figure 2: Read distribution plot across annotation features for any read length.</div></br>

## Count and plot reads of specific lengths

To determine the approximate read length of ribosome footprints in Ribo-Seq experiments, one 
can generate and plot the feature counts for specific read lengths separately. Typically, the 
most abundant read length obtained for translated features corresponds to the approximate footprint 
length occupied by the ribosomes.

```{r featuretypeCounts_length, eval=FALSE}
fc2 <- featuretypeCounts(bfl=BamFileList(outpaths(args), yieldSize=50000), grl=feat, 
                         singleEnd=TRUE, readlength=c(74:76,99:102), type="data.frame")
p2 <- plotfeaturetypeCounts(x=fc2, graphicsfile="results/featureCounts2.png", graphicsformat="png", 
                            scales="fixed", anyreadlength=FALSE, scale_length_val=NULL)
```

![](results/featureCounts2.png)
<div align="center">Figure 3: Read distribution plot across annotation features for specific read lengths.</div></br>


# Adding custom features to workflow

## Predicting uORFs in 5' UTR regions

The function `predORF` can be used to identify open reading frames
(ORFs) and coding sequences (CDSs) in DNA sequences provided as
`DNAString` or `DNAStringSet` objects. The setting
`mode='ORF'` returns continuous reading frames that begin with a start
codon and end with a stop codon, while `mode='CDS'` returns continuous
reading frames that do not need to begin or end with start or stop codons,
respectively. Non-canonical start and stop condons are supported by allowing the
user to provide any custom set of triplets under the `startcodon` and `stopcodon`
arguments (`i.e.` non-ATG start codons). The argument `n` defines the maximum number of ORFs to return for each
input sequence (_e.g._ `n=1` returns only the longest ORF). It also
supports the identification of overlapping and nested ORFs. Alternatively, one 
can return all non-overlapping ORFs including the longest ORF for each input
sequence with `n="all"` and `longest_disjoint=TRUE`.

```{r pred_orf, eval=FALSE}
library(systemPipeRdata); library(GenomicFeatures); library(rtracklayer)
txdb <- makeTxDbFromGFF(file="data/tair10.gff", format="gff3", organism="Arabidopsis")
futr <- fiveUTRsByTranscript(txdb, use.names=TRUE)
dna <- extractTranscriptSeqs(FaFile("data/tair10.fasta"), futr)
uorf <- predORF(dna, n="all", mode="orf", longest_disjoint=TRUE, strand="sense")
```

To use the predicted ORF ranges for expression analysis given genome alignments
as input, it is necessary to scale them to the corresponding genome
coordinates. The function `scaleRanges` does this by transforming the
mappings of spliced features (query ranges) to their corresponding genome
coordinates (subject ranges). The method accounts for introns in the subject
ranges that are absent in the query ranges. The above uORFs predicted in the
provided 5' UTRs sequences using `predORF` are a typical use case
for this application. These query ranges are given relative to the 5' UTR 
sequences and `scaleRanges` will convert them to the corresponding 
genome coordinates. The resulting `GRangesList` object (here `grl_scaled`) 
can be directly used for read counting.

```{r scale_ranges, eval=FALSE}
grl_scaled <- scaleRanges(subject=futr, query=uorf, type="uORF", verbose=TRUE)
export.gff3(unlist(grl_scaled), "results/uorf.gff")
```

To confirm the correctness of the obtained uORF ranges, one can parse their
corresponding DNA sequences from the reference genome with the `getSeq`
function and then translate them with the `translate` function into
proteins. Typically, the returned protein sequences should start with a
`M` (corresponding to start codon) and end with `*`
(corresponding to stop codon). The following example does this for a single uORF 
containing three exons.

```{r translate1, eval=FALSE}
translate(unlist(getSeq(FaFile("data/tair10.fasta"), grl_scaled[[7]])))
```

## Adding custom features to other feature types

If required custom feature ranges can be added to the standard features
generated with the `genFeatures` function above. The following does this for the uORF ranges
predicted with `predORF`.

```{r add_features, eval=FALSE}
feat <- genFeatures(txdb, featuretype="all", reduce_ranges=FALSE)
feat <- c(feat, GRangesList("uORF"=unlist(grl_scaled)))
```

## Predicting sORFs in intergenic regions
The following identifies continuous ORFs in intergenic regions. Note,
`predORF` can only identify continuous ORFs in query sequences. The
function does not identify and remove introns prior to the ORF prediction.  

```{r pred_sorf, eval=FALSE}
feat <- genFeatures(txdb, featuretype="intergenic", reduce_ranges=TRUE)
intergenic <- feat$intergenic
strand(intergenic) <- "+"
dna <- getSeq(FaFile("data/tair10.fasta"), intergenic)
names(dna) <- mcols(intergenic)$feature_by
sorf <- predORF(dna, n="all", mode="orf", longest_disjoint=TRUE, strand="both")
sorf <- sorf[width(sorf) > 60] # Remove sORFs below length cutoff, here 60bp
intergenic <- split(intergenic, mcols(intergenic)$feature_by)
grl_scaled_intergenic <- scaleRanges(subject=intergenic, query=sorf, type="sORF", verbose=TRUE)
export.gff3(unlist(grl_scaled_intergenic), "sorf.gff")
translate(getSeq(FaFile("data/tair10.fasta"), unlist(grl_scaled_intergenic)))
```

# Genomic read coverage along transripts or CDSs
The `featureCoverage` function computes the read coverage along
single and multi component features based on genomic alignments. The coverage
segments of component features are spliced to continuous ranges, such as exons
to transcripts or CDSs to ORFs. The results can be obtained with single
nucleotide resolution (_e.g._ around start and stop codons) or as mean coverage
of relative bin sizes, such as 100 bins for each feature. The latter allows
comparisons of coverage trends among transcripts of variable length. Additionally, 
the results can be obtained for single or many features (_e.g._ any number of
transcripts) at once. Visualization of the coverage results is facilitated by
the downstream `plotfeatureCoverage` function. 

## Binned CDS coverage to compare many transcripts

```{r coverage_binned1, eval=FALSE}
grl <- cdsBy(txdb, "tx", use.names=TRUE)
fcov <- featureCoverage(bfl=BamFileList(outpaths(args)[1:2]), grl=grl[1:4], resizereads=NULL, 
                         readlengthrange=NULL, Nbins=20, method=mean, fixedmatrix=FALSE, 
                         resizefeatures=TRUE, upstream=20, downstream=20,
                         outfile="results/featureCoverage.xls", overwrite=TRUE)
```

## Coverage upstream and downstream of start and stop codons

```{r coverage_binned2, eval=FALSE}
fcov <- featureCoverage(bfl=BamFileList(outpaths(args)[1:4]), grl=grl[1:12], resizereads=NULL, 
                         readlengthrange=NULL, Nbins=NULL, method=mean, fixedmatrix=TRUE, 
                         resizefeatures=TRUE, upstream=20, downstream=20, 
                         outfile="results/featureCoverage.xls", overwrite=TRUE)
plotfeatureCoverage(covMA=fcov, method=mean, scales="fixed", extendylim=2, scale_count_val=10^6)
```

## Combined coverage for both binned CDS and start/stop codons

```{r coverage_binned3, eval=FALSE}
library(ggplot2); library(grid)
fcov <- featureCoverage(bfl=BamFileList(outpaths(args)[1:4]), grl=grl[1:4], resizereads=NULL, 
                         readlengthrange=NULL, Nbins=20, method=mean, fixedmatrix=TRUE, 
                         resizefeatures=TRUE, upstream=20, downstream=20,
                         outfile="results/featureCoverage.xls", overwrite=TRUE)
png("./results/featurePlot.png", height=12, width=24, units="in", res=72)
plotfeatureCoverage(covMA=fcov, method=mean, scales="fixed", extendylim=2, scale_count_val=10^6)
dev.off()
```

![](results/featurePlot.png)
<div align="center">Figure 4: Feature coverage plot with single nucleotide resolution around start and stop codons and binned coverage between them.</div></br>


## Nucleotide level coverage along entire transcripts/CDSs

```{r coverage_nuc_level, eval=FALSE}
fcov <- featureCoverage(bfl=BamFileList(outpaths(args)[1:2]), grl=grl[1], resizereads=NULL, 
                        readlengthrange=NULL, Nbins=NULL, method=mean, fixedmatrix=FALSE, 
                        resizefeatures=TRUE, upstream=20, downstream=20, outfile=NULL)
```

# Read quantification per annotation range

## Read counting with `summarizeOverlaps` in parallel mode using multiple cores

Reads overlapping with annotation ranges of interest are counted for each
sample using the `summarizeOverlaps` function [@Lawrence2013-kt]. The
read counting is preformed for exonic gene regions in a non-strand-specific
manner while ignoring overlaps among different genes. Subsequently, the
expression count values are normalized by \textit{reads per kp per million
mapped reads} (RPKM). The raw read count table (`countDFeByg.xls`) and the correspoding
RPKM table (`rpkmDFeByg.xls`) are written to
separate files in the `results` directory of this project.
Parallelization is achieved with the `BiocParallel` package, here
using 8 CPU cores.

```{r read_counting, eval=FALSE}
library("GenomicFeatures"); library(BiocParallel)
txdb <- loadDb("./data/tair10.sqlite")
eByg <- exonsBy(txdb, by=c("gene"))
bfl <- BamFileList(outpaths(args), yieldSize=50000, index=character())
multicoreParam <- MulticoreParam(workers=8); register(multicoreParam); registered()
counteByg <- bplapply(bfl, function(x) summarizeOverlaps(eByg, x, mode="Union", 
                                               ignore.strand=TRUE, 
                                               inter.feature=FALSE, 
                                               singleEnd=TRUE)) 
countDFeByg <- sapply(seq(along=counteByg), function(x) assays(counteByg[[x]])$counts)
rownames(countDFeByg) <- names(rowRanges(counteByg[[1]])); colnames(countDFeByg) <- names(bfl)
rpkmDFeByg <- apply(countDFeByg, 2, function(x) returnRPKM(counts=x, ranges=eByg))
write.table(countDFeByg, "results/countDFeByg.xls", col.names=NA, quote=FALSE, sep="\t")
write.table(rpkmDFeByg, "results/rpkmDFeByg.xls", col.names=NA, quote=FALSE, sep="\t")
```

Sample of data slice of count table

```{r read_counting_view, eval=FALSE}
read.delim("results/countDFeByg.xls", row.names=1, check.names=FALSE)[1:4,1:5]
```

Sample of data slice of RPKM table

```{r read_rpkm_view, eval=FALSE}
read.delim("results/rpkmDFeByg.xls", row.names=1, check.names=FALSE)[1:4,1:4]
```

Note, for most statistical differential expression or abundance analysis
methods, such as `edgeR` or `DESeq2`, the raw count values
should be used as input. The usage of RPKM values should be restricted to
specialty applications required by some users, _e.g._ manually comparing
the expression levels among different genes or features. 

## Sample-wise correlation analysis

The following computes the sample-wise Spearman correlation coefficients from
the `rlog` transformed expression values generated with the
`DESeq2` package. After transformation to a distance matrix,
hierarchical clustering is performed with the `hclust` function and
the result is plotted as a dendrogram and written to a file named `sample_tree.png`
in the `results` directory. 

```{r sample_tree, eval=FALSE}
library(DESeq2, quietly=TRUE); library(ape,  warn.conflicts=FALSE)
countDF <- as.matrix(read.table("./results/countDFeByg.xls"))
colData <- data.frame(row.names=targetsin(args)$SampleName, condition=targetsin(args)$Factor)
dds <- DESeqDataSetFromMatrix(countData = countDF, colData = colData, design = ~ condition)
d <- cor(assay(rlog(dds)), method="spearman")
hc <- hclust(dist(1-d))
png("results/sample_tree.pdf")
plot.phylo(as.phylo(hc), type="p", edge.col="blue", edge.width=2, show.node.label=TRUE, 
           no.margin=TRUE)
dev.off()
```

![](results/sample_tree.png)
<div align="center">Figure 5: Correlation dendrogram of samples.</div></br>


# Analysis of differentially expressed genes with `edgeR`

The analysis of differentially expressed genes (DEGs) is performed with the glm
method from the `edgeR` package [@Robinson2010-uk]. The sample
comparisons used by this analysis are defined in the header lines of the
`targets.txt` file starting with `<CMP>`.

```{r deg_edger, eval=FALSE}
library(edgeR)
countDF <- read.delim("results/countDFeByg.xls", row.names=1, check.names=FALSE) 
targets <- read.delim("targets.txt", comment="#")
cmp <- readComp(file="targets.txt", format="matrix", delim="-")
edgeDF <- run_edgeR(countDF=countDF, targets=targets, cmp=cmp[[1]], independent=FALSE, mdsplot="")
```

Add functional gene descriptions, here from `biomaRt`. 

```{r add_descr, eval=FALSE}
library("biomaRt")
m <- useMart("plants_mart", dataset="athaliana_eg_gene", host="plants.ensembl.org")
desc <- getBM(attributes=c("tair_locus", "description"), mart=m)
desc <- desc[!duplicated(desc[,1]),]
descv <- as.character(desc[,2]); names(descv) <- as.character(desc[,1])
edgeDF <- data.frame(edgeDF, Desc=descv[rownames(edgeDF)], check.names=FALSE)
write.table(edgeDF, "./results/edgeRglm_allcomp.xls", quote=FALSE, sep="\t", col.names = NA)
```

Filter and plot DEG results for up and down regulated genes. The definition of
`up` and `down` is given in the corresponding help file. To
open it, type `?filterDEGs` in the R console. 

```{r filter_degs, eval=FALSE}
edgeDF <- read.delim("results/edgeRglm_allcomp.xls", row.names=1, check.names=FALSE) 
png("./results/DEGcounts.png", height=10, width=10, units="in", res=72)
DEG_list <- filterDEGs(degDF=edgeDF, filter=c(Fold=2, FDR=20))
dev.off()
write.table(DEG_list$Summary, "./results/DEGcounts.xls", quote=FALSE, sep="\t", row.names=FALSE)
```

![](results//DEGcounts.png)
<div align="center">Figure 6: Up and down regulated DEGs.</div></br>

The function `overLapper` can compute Venn intersects for large
numbers of sample sets (up to 20 or more) and `vennPlot` can plot 2-5
way Venn diagrams. A useful feature is the possiblity to combine the counts
from several Venn comparisons with the same number of sample sets in a single
Venn diagram (here for 4 up and down DEG sets).

```{r venn_diagram, eval=FALSE}
vennsetup <- overLapper(DEG_list$Up[6:9], type="vennsets")
vennsetdown <- overLapper(DEG_list$Down[6:9], type="vennsets")
png("results/vennplot.png")
vennPlot(list(vennsetup, vennsetdown), mymain="", mysub="", colmode=2, ccol=c("blue", "red"))
dev.off()
```

![](results/vennplot.png)
<div align="center">Figure 7: Venn Diagram for 4 Up and Down DEG Sets</div></br>

# GO term enrichment analysis of DEGs}

## Obtain gene-to-GO mappings
The following shows how to obtain gene-to-GO mappings from `biomaRt`
(here for _A. thaliana_) and how to organize them for the downstream GO
term enrichment analysis. Alternatively, the gene-to-GO mappings can be
obtained for many organisms from Bioconductor's `*.db` genome
annotation packages or GO annotation files provided by various genome
databases. For each annotation this relatively slow preprocessing step needs to
be performed only once. Subsequently, the preprocessed data can be loaded with
the `load` function as shown in the next subsection. 

```{r get_go_annot, eval=FALSE}
library("biomaRt")
listMarts() # To choose BioMart database
listMarts(host="plants.ensembl.org")
m <- useMart("plants_mart", host="plants.ensembl.org")
listDatasets(m)
m <- useMart("plants_mart", dataset="athaliana_eg_gene", host="plants.ensembl.org")
listAttributes(m) # Choose data types you want to download
go <- getBM(attributes=c("go_accession", "tair_locus", "go_namespace_1003"), mart=m)
go <- go[go[,3]!="",]; go[,3] <- as.character(go[,3])
go[go[,3]=="molecular_function", 3] <- "F"; go[go[,3]=="biological_process", 3] <- "P"; go[go[,3]=="cellular_component", 3] <- "C"
go[1:4,]
dir.create("./data/GO")
write.table(go, "data/GO/GOannotationsBiomart_mod.txt", quote=FALSE, row.names=FALSE, col.names=FALSE, sep="\t")
catdb <- makeCATdb(myfile="data/GO/GOannotationsBiomart_mod.txt", lib=NULL, org="", colno=c(1,2,3), idconv=NULL)
save(catdb, file="data/GO/catdb.RData")
```

## Batch GO term enrichment analysis

Apply the enrichment analysis to the DEG sets obtained the above differential
expression analysis. Note, in the following example the `FDR` filter is set
here to an unreasonably high value, simply because of the small size of the toy
data set used in this vignette. Batch enrichment analysis of many gene sets is
performed with the function. When `method=all`, it returns all GO terms passing
the p-value cutoff specified under the `cutoff` arguments. When `method=slim`,
it returns only the GO terms specified under the `myslimv` argument. The given
example shows how a GO slim vector for a specific organism can be obtained from
BioMart.

```{r go_enrich, eval=FALSE, warning=FALSE, message=FALSE}
library("biomaRt")
library(BBmisc) # Defines suppressAll()
load("data/GO/catdb.RData")
DEG_list <- filterDEGs(degDF=edgeDF, filter=c(Fold=2, FDR=50), plot=FALSE)
up_down <- DEG_list$UporDown; names(up_down) <- paste(names(up_down), "_up_down", sep="")
up <- DEG_list$Up; names(up) <- paste(names(up), "_up", sep="")
down <- DEG_list$Down; names(down) <- paste(names(down), "_down", sep="")
DEGlist <- c(up_down, up, down)
DEGlist <- DEGlist[sapply(DEGlist, length) > 0]
BatchResult <- GOCluster_Report(catdb=catdb, setlist=DEGlist, method="all", id_type="gene", CLSZ=2, cutoff=0.9, gocats=c("MF", "BP", "CC"), recordSpecGO=NULL)
library("biomaRt")
m <- useMart("plants_mart", dataset="athaliana_eg_gene", host="plants.ensembl.org")
goslimvec <- as.character(getBM(attributes=c("goslim_goa_accession"), mart=m)[,1])
BatchResultslim <- GOCluster_Report(catdb=catdb, setlist=DEGlist, method="slim", id_type="gene", myslimv=goslimvec, CLSZ=10, cutoff=0.01, gocats=c("MF", "BP", "CC"), recordSpecGO=NULL)
```

The `data.frame` generated by `GOCluster` can be plotted with the `goBarplot` function. Because of the
variable size of the sample sets, it may not always be desirable to show
the results from different DEG sets in the same bar plot. Plotting
single sample sets is achieved by subsetting the input data frame as
shown in the first line of the following example.

```{r go_plot, eval=FALSE}
gos <- BatchResultslim[grep("M6-V6_up_down", BatchResultslim$CLID), ]
gos <- BatchResultslim
png("./results/GOslimbarplotMF.png", height=12, width=12, units="in", res=72)
goBarplot(gos, gocat="MF")
dev.off()
goBarplot(gos, gocat="BP")
goBarplot(gos, gocat="CC")
```

![](results/GOslimbarplotMF.png)
<div align="center">Figure 8: GO Slim Barplot for MF Ontology.</div></br>

# Differential ribosome loading analysis (translational efficiency)

Combinded with mRNA-Seq data, Ribo-Seq or polyRibo-Seq experiments can be used
to study changes in translational efficiencies of genes and/or transcripts for
different treatments. For test purposes the following generates a small test
data set from the sample data used in this vignette, where two types of RNA
samples (`assays`) are considered: polyribosomal mRNA (`Ribo`)
and total mRNA (`mRNA`). In addition, there are two treatments
(`conditions`): `M1` and `A1`. 

```{r diff_loading, eval=TRUE}
library(DESeq2)
targetspath <- system.file("extdata", "targetsPE.txt", package="systemPipeR")
parampath <- system.file("extdata", "tophat.param", package="systemPipeR")
countDFeBygpath <- system.file("extdata", "countDFeByg.xls", package="systemPipeR")
args <- suppressWarnings(systemArgs(sysma=parampath, mytargets=targetspath))
countDFeByg <- read.delim(countDFeBygpath, row.names=1)
coldata <- DataFrame(assay=factor(rep(c("Ribo","mRNA"), each=4)), 
                condition=factor(rep(as.character(targetsin(args)$Factor[1:4]), 2)), 
                row.names=as.character(targetsin(args)$SampleName)[1:8])
coldata
```

Differences in translational efficiencies can be calculated by ratios of ratios
for the two conditions: 

$$(Ribo\_A1 / mRNA\_A1) / (Ribo\_M1 / mRNA\_M1)$$


The latter can be modeled with the `DESeq2` package using the design $\sim assay + condition + assay:condition$, 
where the interaction term $assay:condition$
represents the ratio of ratios. Using the likelihood ratio test of
`DESeq2`, which removes the interaction term in the reduced model, one
can test whether the translational efficiency (ribosome loading) is different
in condition `A1` than in `M1`.

```{r diff_translational_eff, eval=TRUE}
dds <- DESeqDataSetFromMatrix(countData=as.matrix(countDFeByg[,rownames(coldata)]), 
                            colData = coldata, 
                            design = ~ assay + condition + assay:condition)
# model.matrix(~ assay + condition + assay:condition, coldata) # Corresponding design matrix
dds <- DESeq(dds, test="LRT", reduced = ~ assay + condition)
res <- DESeq2::results(dds)
head(res[order(res$padj),],4)
# write.table(res, file="transleff.xls", quote=FALSE, col.names = NA, sep="\t")
```

# Clustering and heat maps

The following example performs hierarchical clustering on the `rlog` transformed expression matrix subsetted by the DEGs identified in the 
above differential expression analysis. It uses a Pearson correlation-based distance measure and complete linkage for cluster joining.

```{r heatmap, eval=FALSE}
library(pheatmap)
geneids <- unique(as.character(unlist(DEG_list[[1]])))
y <- assay(rlog(dds))[geneids, ]
png("heatmap1.png")
pheatmap(y, scale="row", clustering_distance_rows="correlation", clustering_distance_cols="correlation")
dev.off()
```

![](results/heatmap1.png)
<div align="center">Figure 9: Heat map with hierarchical clustering dendrograms of DEGs</div></br>

# Render report in HTML and PDF format

```{r render_report, eval=FALSE}
rmarkdown::render("systemPipeRIBOseq.Rmd", "html_document")
rmarkdown::render("systemPipeRIBOseq.Rmd", "pdf_document")
```


# Version Information

```{r sessionInfo}
sessionInfo()
```

# Funding
This research was funded by National Science Foundation Grants IOS-0750811 and
MCB-1021969, and a Marie Curie European Economic Community Fellowship
PIOF-GA-2012-327954.

# References