This vignette will cover a wide range of analytical and visualization techniques involved in a typical pathway analysis. The Overview section will go into more detail on the particulars, but note that this vignette is designed to be modular and carefully considered. Please do not simply run the entire script and expect to get anything meaningful from the final output. This is an instructional device, ideal for guided workshops.

1 Installation

First, make sure you have rWikiPathways installed…

if(!"rWikiPathways" %in% installed.packages()){
    if (!requireNamespace("BiocManager", quietly = TRUE))
        install.packages("BiocManager")
    BiocManager::install("rWikiPathways", update = FALSE)
}
library(rWikiPathways)

We will be using a diverse set of R packages in this vignette. This next chunk should handle the installation and loading of all the R packages we will need. The final printed line will state whether it was successful or not. Good luck!

As with most installations, you may be prompted for responses. For this vignette, you can reply ‘a’ to update all old packages (if prompted) and ‘no’ to skip compiling from source (if an option for your setup). Please verify the final “success”" message printed at the end before proceeding.

load.libs <- c(
  "DOSE",
  "GO.db",
  "GSEABase",
  "org.Hs.eg.db",
  "clusterProfiler",
  "dplyr",
  "tidyr",
  "ggplot2",
  "stringr",
  "RColorBrewer",
  "rWikiPathways",
  "RCy3")
options(install.packages.check.source = "no")
options(install.packages.compile.from.source = "never")
if (!require("pacman")) install.packages("pacman"); library(pacman)
p_load(load.libs, update = TRUE, character.only = TRUE)
status <- sapply(load.libs,require,character.only = TRUE)
if(all(status)){
    print("SUCCESS: You have successfully installed and loaded all required libraries.")
} else{
    cat("ERROR: One or more libraries failed to install correctly. Check the following list for FALSE cases and try again...\n\n")
    status
}

The RCy3 package is used to connect with Cytoscape. So you will also need to install and launch Cytoscape:

cytoscapePing()  #this will tell you if you're able to successfully connect to Cytoscape or not

For this vignette, you’ll also need a couple apps for Cytoscape. With Cytoscape running, you can install each of these from the Cytoscape App Store with a single click:

If you are running Cytoscape 3.7.0 or above, you can simply run these commands:

installApp('WikiPathways') 
installApp('CyTargetLinker') 
installApp('stringApp') 

2 Overview

In this vignette, we will be performing functional enrichment analysis on a differential gene expression dataset. The dataset compares the expression of transcripts in lung cancer biopses versus normal tissue. Differential expression analysis has already been performed, generating log2foldchange and P-values for each gene. The enrichment analysis will be performed against Gene Ontology, as an introduction to the most common type of enrichment, commonly referred to as GO Analysis. This will serve as the foundation for more advanced enrichment analysis against a pathway database, which is called Pathway Analysis.

Working with pathways opens up unique analysis and visualization options. We will query WikiPathways for relevant content and import pathway models into Cytoscape. In Cytoscape, we will perform data overlays, add drug interactions and generate high-quality images for publication.

3 Dataset

The format of this lung cancer dataset should look familiar to anyone who has worked with differential gene expression results. It contains columns of gene identifiers (Ensembl IDs), gene symbols, log2foldchange values, P-Values and adjusted P-Values.

lung.expr <- read.csv(system.file("extdata","data-lung-cancer.csv", package="rWikiPathways"),stringsAsFactors = FALSE)
nrow(lung.expr)
head(lung.expr)

Now let’s prepare up- and down-regulated gene lists using some conventional criteria.

up.genes <- lung.expr[lung.expr$log2FC > 1 & lung.expr$adj.P.Value < 0.05, 1] 
dn.genes <- lung.expr[lung.expr$log2FC < -1 & lung.expr$adj.P.Value < 0.05, 1]
bkgd.genes <- lung.expr[,1]

4 Enrichment

With our gene sets in hand, we are ready to perform enrichment analysis… well, almost. Typical of any bioinformatics analysis, we need to be aware of which gene identifiers we are dealing with. We have Ensembl IDs, but the package we are going to use wants Entrez IDs. Fortunately, the package provides its own converter called bitr. This is the first function in the clusterProfiler package that we’ll be using:

up.genes.entrez <- clusterProfiler::bitr(up.genes,fromType = "ENSEMBL",toType = "ENTREZID",OrgDb = org.Hs.eg.db)
cat("\n\nWhich column contains my new Entrez IDs?\n")
head(up.genes.entrez)

Note that conversions are rarely 100% complete (e.g., due to one-to-many mappings), so this tool reports what it failed to convert. And now we have a dataframe with a new column of Entrez IDs paired with our original list of Ensembl IDs.

Here’s the complete list of identifiers that this particular tool can convert across. You have to spell these precisely and in all caps for the bitr function to work:

keytypes(org.Hs.eg.db)

Let’s convert our other lists to Entrez IDs:

dn.genes.entrez <- bitr(dn.genes,fromType = "ENSEMBL",toType = "ENTREZID",OrgDb = org.Hs.eg.db)
bkgd.genes.entrez <- bitr(bkgd.genes,fromType = "ENSEMBL",toType = "ENTREZID",OrgDb = org.Hs.eg.db)

4.1 Gene Ontology

Ok. Now we are ready to perform enrichment analysis. Let’s start with Gene Ontology. Note: This can take up to a miunte to run… because there are so many GO terms!

egobp <- clusterProfiler::enrichGO(
        gene     = up.genes.entrez[[2]],
        universe = bkgd.genes.entrez[[2]],
        OrgDb    = org.Hs.eg.db,
        ont      = "BP",
        pAdjustMethod = "fdr",
        pvalueCutoff = 0.05, #p.adjust cutoff (https://github.com/GuangchuangYu/clusterProfiler/issues/104)
        readable = TRUE)

head(egobp,10)

While it’s running, you can examine the parameters and their meaning…

It’s done! Does the format and results make sense? Do you understand the two Ratio columns? Does the biology of the top results make sense?

Tables are fine, but we’re in R to see plots. Conveniently, clusterProfiler provides a variety of plots with default settings:

barplot(egobp, showCategory = 20)
dotplot(egobp, showCategory = 20)
emapplot(egobp, showCategory = 20)
goplot(egobp)

Check out the clusterProfiler vignette for other supported functions and plots.

For more control and customization, here is an example using ggplot and the same enrichGO object output from clusterProfiler

ggplot(egobp[1:20], aes(x=reorder(Description, -pvalue), y=Count, fill=-p.adjust)) +
    geom_bar(stat = "identity") +
    coord_flip() +
    scale_fill_continuous(low="blue", high="red") +
    labs(x = "", y = "", fill = "p.adjust") +
    theme(axis.text=element_text(size=11))

Use str(egobp) to explore the data structure in more detail to see what you can use in plotting.

4.2 WikiPathways

That’s all well and fine, but we are here for Pathway Analysis! Everything up to this point is basic GO analysis. Building on top of this foundation, let’s see what WikiPathways can add. The clusterProfiler package doesn’t include built-in support for WikiPathways (yet), but it does support the GMT standard file format and so does WikiPathways! It just takes a few commands to prepare WikiPathways data for clusterProfiler’s enricher function, which is the generic analog of enrichGO.

The first step is to retreive the latest GMT file listing all the genes in each of the human pathways. WikiPathways releases a GMT file per species every month and archives them for reproducible research. Here, we’ll use the first function from the rWikiPathways package, aptly named downloadPathwayArchive

wp.hs.gmt <- rWikiPathways::downloadPathwayArchive(organism="Homo sapiens", format = "gmt")

# supported organisms (and case-dependent spelling)...
listOrganisms()

Now that we’ve got the latest GMT file for human pathways, we can feed it into clusterProfiler and generate the two dataframes we need for enricher

wp2gene <- clusterProfiler::read.gmt(wp.hs.gmt)
wp2gene <- wp2gene %>% tidyr::separate(ont, c("name","version","wpid","org"), "%")
wpid2gene <- wp2gene %>% dplyr::select(wpid,gene) #TERM2GENE
wpid2name <- wp2gene %>% dplyr::select(wpid,name) #TERM2NAME
wpid2gene
wpid2name

That’s it! We are ready to perform pathway analysis…

ewp.up <- clusterProfiler::enricher(
        up.genes.entrez[[2]],
        universe = bkgd.genes.entrez[[2]],
        pAdjustMethod = "fdr",
        pvalueCutoff = 0.1, #p.adjust cutoff; relaxed for demo purposes
        TERM2GENE = wpid2gene,
        TERM2NAME = wpid2name)

head(ewp.up)

For some reason, enricher doesn’t automatically add gene symbols to the result object, but there is a handy function in DOSE that does…

ewp.up <- DOSE::setReadable(ewp.up, org.Hs.eg.db, keytype = "ENTREZID")
head(ewp.up)

And, we have access to all the same plotting functions as before…

barplot(ewp.up, showCategory = 20)
dotplot(ewp.up, showCategory = 20)
emapplot(ewp.up, showCategory = 20)

Before we forget, we can also do the same analysis for down-regulated genes…

ewp.dn <- enricher(
  dn.genes.entrez[[2]],
  #universe = bkgd.genes[[2]],  #hint: comment out to get any results for demo
  pAdjustMethod = "fdr",
  pvalueCutoff = 0.1, #p.adjust cutoff; relaxed for demo purposes
  TERM2GENE = wpid2gene,
  TERM2NAME = wpid2name)

 ewp.dn <- setReadable(ewp.dn, org.Hs.eg.db, keytype = "ENTREZID")
 head(ewp.dn)
 dotplot(ewp.dn, showCategory = 20)

Interesting… Almost twice as many down-regulated genes (634 vs 366), but fewer significant pathway hits. Pathway analysis is a more focused approach than GO analysis. It requires sets of genes that are functionally related in the context of known pathways mechanisms. So, gene set size doesn’t always correlate with result size.

BONUS: What we just did is also refered to as Over-representation Analysis (ORA). Another approach is Gene Set Enrichment Analysis (GSEA). One advantage of GSEA is that you don’t have to pick an arbitrary log2FC cutoff to define gene sets. Instead you provide a pre-ordered list of significant gene-named log2FC values at let the GSEA do the work. Here are the steps, see if you can follow along:

lung.expr$fcsign <- sign(lung.expr$log2FC)
lung.expr$logfdr <- -log10(lung.expr$P.Value)
lung.expr$sig <- lung.expr$logfdr/lung.expr$fcsign
sig.lung.expr.entrez<-merge(lung.expr, bkgd.genes.entrez, by.x = "GeneID", by.y = "ENSEMBL")
gsea.sig.lung.expr <- sig.lung.expr.entrez[,8]
names(gsea.sig.lung.expr) <- as.character(sig.lung.expr.entrez[,9])
gsea.sig.lung.expr <- sort(gsea.sig.lung.expr,decreasing = TRUE)

gwp.sig.lung.expr <- clusterProfiler::GSEA(
  gsea.sig.lung.expr,
  pAdjustMethod = "fdr",
  pvalueCutoff = 0.05, #p.adjust cutoff
  TERM2GENE = wpid2gene,
  TERM2NAME = wpid2name)

gwp.sig.lung.expr.df = data.frame(ID=gwp.sig.lung.expr$ID,
                                 Description=gwp.sig.lung.expr$Description,
                                 enrichmentScore=gwp.sig.lung.expr$enrichmentScore,
                                 NES=gwp.sig.lung.expr$NES,
                                 pvalue=gwp.sig.lung.expr$pvalue,
                                 p.adjust=gwp.sig.lung.expr$p.adjust,
                                 rank=gwp.sig.lung.expr$rank,
                                 leading_edge=gwp.sig.lung.expr$leading_edge
)
gwp.sig.lung.expr.df[which(gwp.sig.lung.expr.df$NES > 1),] #pathways enriched for upregulated lung cancer genes
gwp.sig.lung.expr.df[which(gwp.sig.lung.expr.df$NES < -1),] #pathways enriched for downregulated lung cancer genes

One of the advantages of pathway analysis is that you have pathway models already built and ready for data overlays. We will get to that soon in the in Visualize section, but first, let’s see what else we can learn about the pathways hits from WikiPathways.

5 Explore

Turning again to the rWikiPathways package, let’s explore the content and some of our pathway hits so far. Since we are studying Lung Cancer here, let’s start with a search for relevant pathways…

findPathwayNamesByText("lung cancer")

Whoa, that’s a lot… and there’s a bunch of repeats!? This general search includes all matches to “lung” and/or “cancer”, sorted by best matches to both terms. And it also includes matches from all speceis, e.g., mouse and rat, in addition to human. Let’s be more specific…

lc.pathways <- findPathwaysByText('"lung cancer"')  #quotes inside query to require both terms
human.lc.pathways <- lc.pathways %>% 
  dplyr::filter(species == "Homo sapiens") # just the human lung cancer pathways
human.lc.pathways$name # display the pathway titles

Ok, so there are just a few human pathways that explicitly mention “lung cancer” in their titles or descriptions. None of these were in our top hits for enrichment, but we might want to look at them anyways during our exploratory data visualization, right? So, let hold on to their WPIDs for now…

lc.wpids <- human.lc.pathways$id
lc.wpids

You can also search pathways by gene identifiers, pubmed references and ontology terms. But we already know the primary pathways we want to see based on our pathway enrichment analysis. Let’s identify their WPIDs…

ewp.up.wpids <- ewp.up$ID
ewp.up.wpids

Let’s take a look at these. We could open them in our browser for example…

url <- getPathwayInfo("WP179")$url
browseURL(url)

You can access all of the WikiPathways website information using rWikiPathways. You can even query the history of a particular pathway or recent changes across the entire site. It’s a wiki after all!

But what we really want to do next is view our data on these pathways. For that, we are going to turn to Cytoscape and the RCy3 package.

6 Visualize

Cytoscape is a popular network visualization and analysis tool with great community support for development and scripting. Since pathways are just a special type of network, it’s perfect for providing high quality visualization for pathway analysis results.

We have the RCy3 package loaded already, but we also have to have Cytoscape launched (see step 1. Installation, if you haven’t already). Once Cytoscape is running, try to ping it with this command:

cytoscapePing()

If you’ve got everything loaded and running, then all you need to do is run this command to import a pathway into Cytoscape from WikiPathways:

RCy3::commandsRun('wikipathways import-as-pathway id=WP179') 

There it is! The latest approved version of the pathway, now in Cytoscape as a network model with annotated genes, proteins and metabolites. For performance reasons, Cytoscape sets a view threshold to hide details (like node labels) when zoomed out. If you want to override this, use…

toggleGraphicsDetails()

Let’s load the same data we used in the enrichment analysis that pointed us to this pathway in the first place. We’ll just need to tell Cytoscape which column in our data contains identifiers (in this case Ensembl IDs) and this column in the Cytoscape Node Table contains corresponding identifiers.

loadTableData(lung.expr, data.key.column = "GeneID", table.key.column = "Ensembl")

Now we can define visual styles to visualize our data on this pathway. First, let’s set the node fill color to display the log2 fold change data. We will need to supply a range of data values and a matching range of colors. For the range of data values, let’s identify the largest min or max value in our dataset:

min.lung.expr = min(lung.expr["log2FC"],na.rm=TRUE)
max.lung.expr = max(lung.expr["log2FC"],na.rm=TRUE)
abs.lung.expr = max(abs(min.lung.expr),max.lung.expr)
data.values = c(-abs.lung.expr,0,abs.lung.expr)

Next, we use the RColorBrewer package to help us pick good colors to pair with our data values.

display.brewer.all(length(data.values), colorblindFriendly=TRUE, type="div") # div,qual,seq,all
node.colors <- c(rev(brewer.pal(length(data.values), "RdBu")))

Now we are ready to set the node fill color…

setNodeColorMapping("log2FC", data.values, node.colors, default.color = "#FFFFFF", style.name = "WikiPathways")

Alternatively, we could just take the max value actually mapped to our pathway:

data.values<-c(-5,0,5)
setNodeColorMapping("log2FC", data.values, node.colors, default.color = "#FFFFFF", style.name = "WikiPathways")

You can similarly map P-values to border color, etc. There are dozens and dozens of visual properties on nodes and edges availble for data visualization!

The power of scripting is in doing something multiple times though… So, let’s now apply this same data and visual style to all the pathways we are interested in. In just two lines of code…

lapply(ewp.up.wpids, function (x) {
    commandsRun(paste0('wikipathways import-as-pathway id=',x))
    loadTableData(lung.expr, data.key.column = "GeneID", table.key.column = "Ensembl")
    toggleGraphicsDetails()
    })

BONUS: What about those lung cancer pathways we found? Let’s take a look at the data overlay on those as well.

lapply(lc.wpids, function (x){
    commandsRun(paste0('wikipathways import-as-pathway id=',x))
    loadTableData(lung.expr, data.key.column = "GeneID", table.key.column = "Ensembl")
    toggleGraphicsDetails()
    })

7 Extend

Now that our pathways are loaded into Cytoscape, this opens up a ton of potential analysis and visualizations options! Check out the Cytoscape manual and App Store, for starters. You might also browse the Cytoscape tutorials and RCy3 vignettes if you want hands-on examples.

In this vignette, we will use the CyTargetLinker app for Cytoscape to extend a network representation of our pathway with drug-target interactions.

First, let’s reimport our pathway as a network using this slightly modified command:

commandsRun('wikipathways import-as-network id=WP179')
loadTableData(lung.expr, data.key.column = "GeneID", table.key.column = "Ensembl") 
setNodeColorMapping("log2FC", data.values, node.colors, default.color = "#FFFFFF", style.name = "WikiPathways-As-Network") 

See the difference? Same data, same pathway source, but different representation. The network view of pathways is useful when you want to add more nodes, traverse paths, perform automatic layouts, etc.

Next, we need to load the latest drug-target database. The databases supported by CyTargetLinker are called linksets and can be downloaded from the CyTargetLinker website. We have provided an example drugbank linkset for this vignette, so you don’t have to download anything.

unzip(system.file("extdata","drugbank-5.1.0.xgmml.zip", package="rWikiPathways"), exdir = getwd())
drugbank <- file.path(getwd(), "drugbank-5.1.0.xgmml")

Now that we have our drugbank linkset loaded, we can run CyTargetLinker as a command:

commandsRun(paste0('cytargetlinker extend idAttribute="Ensembl" linkSetFiles="', drugbank, '"') )
commandsRun('cytargetlinker applyLayout network="current"')

This returns information about what was added. And in Cytoscape, you now have a copy of your original network, but now with additional nodes and edges. Check it out…

…Hmm, they are kind of plain and hard to see. Let’s use Cytoscape visualization styles to fix that!

my.drugs <- selectNodes("drug", by.col = "CTL.Type", preserve = FALSE)$nodes #easy way to collect node SUIDs by column value
clearSelection()
setNodeColorBypass(my.drugs, "#DD99FF")
setNodeShapeBypass(my.drugs, "hexagon")

drug.labels <- getTableColumns(columns=c("SUID","CTL.label"))
drug.labels <- na.omit(drug.labels)
mapply(function(x,y) setNodeLabelBypass(x,y), drug.labels$SUID, drug.labels$CTL.label)

Now we have drugs labeled as purple hexagons interacting with the network view of our WikiPathways hit from our functional enrichment analysis on a lung cancer dataset. Cool! Hopefully, you can imagine now, extending this (or other pathawys) in other ways, e.g., TF and miRNA interactions. Or applying numerous other Cytoscape apps to the analysis and visualization of this result.

8 Save

Last, but not least, be sure to save your work along the way. Here is how you might save your up and down regulated subsets from the lung cancer dataset as R objects that are easy to read back in and share:

save(ewp.up, file = "lung_cancer_ewp_up.Rdata")
save(ewp.dn, file = "lung_cancer_ewp_down.Rdata")

Session files save everything that is in Cytoscape, including the pathways, networks, data and styles. As with most project software, we recommend saving often!

full.path=paste(getwd(),'tutorial_session',sep='/')
saveSession(filename=full.path) #.cys

Note: If you don’t specify a complete path, the files will be save relative to your Cytoscape installation directory, e.g., /Applications/Cytoscape_v3.6.1/… or somewhere you migt not have write permissions.

You can export extremely high resolution images, including vector graphic formats.

full.path=paste(getwd(),'tutorial_image2',sep='/')
exportImage(filename=full.path, type='PDF') #.pdf
exportImage(filename=full.path, type='PNG', zoom=200) #.png; use zoom or width args to increase size/resolution
?exportImage

And with so many libraries involved, it’s a good idea to keep track of the version information when performing real analyses so that you (and others) can reliably reproduce your results:

sessionInfo()