In rosscm/FEDUP: Fisher's Test for Enrichment and Depletion of User-Defined Pathways

knitr::opts_chunk$set(
    collapse = TRUE,
    comment = "#>",
    fig.path = "vignettes/figures/",
    out.width = "100%"
)

This is an R package that tests for enrichment and depletion of user-defined pathways using a Fisher's exact test. The method is designed for versatile pathway annotation formats (eg. gmt, txt, xlsx) to allow the user to run pathway analysis on custom annotations. This package is also integrated with Cytoscape to provide network-based pathway visualization that enhances the interpretability of the results.

This vignette will explain how to use fedup when testing multiple sets of genes for pathway enrichment and depletion.

System prerequisites

R version ≥ 4.1
R packages:

CRAN: openxlsx, tibble, dplyr, data.table, ggplot2, ggthemes, forcats, RColorBrewer
Bioconductor: RCy3

Installation

Install fedup from Bioconductor:

if(!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("fedup")

Or install the development version from Github:

devtools::install_github("rosscm/fedup", quiet = TRUE)

Load necessary packages:

library(fedup)
library(dplyr)
library(tidyr)
library(ggplot2)

Running the package

Input data

Load test genes (geneMulti) and pathways annotations (pathwaysGMT):

data(geneMulti)
data(pathwaysGMT)

Take a look at the data structure:

str(geneMulti)
str(head(pathwaysGMT))

To see more info on this data, run ?geneDouble or ?pathwaysGMT. You could also run example("prepInput", package = "fedup") or example("readPathways", package = "fedup") to see exactly how the data was generated using the prepInput() and readPathways() functions. ? and example() can be used on any other functions mentioned here to see their documentation and run examples.

The sample geneMulti list object contains thirteen vector elements: background, FASN_negative, FASN_positive, ACACA_negative, ACACA_positive, LDLR_negative, LDLR_positive, SREBF1_negative, SREBF1_positive, SREBF2_negative, SREBF2_positive, C12orf49_negative, and C12orf49_positive. The background consists of all genes that the test sets (in this case all sets besides background) will be compared against. FASN_negative consists of genes that form negative genetic interactions with the FASN gene after CRISPR-Cas9 knockout. FASN_positive consists of genes that form positve genetic interactions with FASN. The remaining elements contain genes forming genetic interactions with their respective genes (ACACA, LDLR, SREBF1, SREBF2, C12orf49). If you're interested in seeing how this data set was constructed, check out the code. Also, the paper the data was taken from is found here.

As an example, FASN is a fatty acid synthase, so we would expect to see enrichment of the negative interactions for pathways associated with sensitization of fatty acid synthesis, as well as enrichment of the positive interactions for pathways associated with suppression of the function. Conversely, we expect to find depletion for pathways not at all involved with FASN biology. Let's see!

Pathway analysis

Now use runFedup on the sample data:

fedupRes <- runFedup(geneMulti, pathwaysGMT)

The fedupRes output is a list of length length(which(names(geneMulti) != "background")), corresponding to the number of test sets in geneMulti (i.e., 12).

View fedup results for FASN_negative sorted by pvalue:

set <- "FASN_negative"
print(head(fedupRes[[set]][which(fedupRes[[set]]$status == "enriched"),]))
print(head(fedupRes[[set]][which(fedupRes[[set]]$status == "depleted"),]))

Here we see the strongest enrichment for the ASPARAGINE N-LINKED GLYCOSYLATION pathway. Given that FASN mutant cells show a strong dependence on lipid uptake, this enrichment for negative interactions with genes involved in glycosylation is expected. We also see significant enrichment for other related pathways, including DISEASES ASSOCIATED WITH N-GLYCOSYLATION OF PROTEINS and DISEASES OF GLYCOSYLATION. Conversely, we see significant depletion for functions not associated with these processes, such as OLFACTORY SIGNALING PATHWAY, GPCR LIGAND BINDING and KERATINIZATION. Nice!

Let's also view fedup results for FASN_positive, sorted by pvalue:

set <- "FASN_positive"
print(head(fedupRes[[set]][which(fedupRes[[set]]$status == "enriched"),]))
print(head(fedupRes[[set]][which(fedupRes[[set]]$status == "depleted"),]))

Results for any test set can be indexed by its name:

names(fedupRes)

Dot plot

Prepare data for plotting via dplyr and tidyr:

fedupPlot <- fedupRes %>%
    bind_rows(.id = "set") %>%
    separate(col = "set", into = c("set", "sign"), sep = "_") %>%
    subset(qvalue < 0.05) %>%
    mutate(log10qvalue = -log10(qvalue)) %>%
    mutate(pathway = gsub("\\%.*", "", pathway)) %>%
    as.data.frame()

Since we're dealing with multiple test sets here, it's important we create the set and sign columns in fedupPlot to distinguish between them. Take a look at ?dplyr::bind_rows for details on how the output fedup results list (fedupRes) was bound into a single dataframe and ?tidyr::separate for how the set and sign columns were created.

Plot significant results (qvalue < 0.05) in the form of a dot plot via plotDotPlot. Facet the points by the set and sign columns and colour by sign:

p <- plotDotPlot(
        df = fedupPlot,
        xVar = "log10qvalue",
        yVar = "pathway",
        xLab = "-log10(qvalue)",
        fillVar = "sign",
        fillLab = "Genetic interaction",
        fillCol = c("#6D90CA", "#F6EB13"),
        sizeVar = "fold_enrichment",
        sizeLab = "Fold enrichment") +
    facet_grid("sign ~ set", scales = "free_y", space = "free") +
    theme(strip.text.y = element_blank())
print(p)

Look at all those chick... enrichments! This is a bit overwhelming, isn't it? How do we interpret these 244 fairly redundant pathways in a way that doesn't hurt our tired brains even more? Oh I know, let's first try another ggplot-based plot.

We can instead plot the degree of fold enrichment for a subset of pathways across our test sets. First, select the top 20 results to plot from fedupRes:

topPath <- fedupRes %>%
    bind_rows(.id = "set") %>%
    arrange(desc(fold_enrichment)) %>%
    slice(1:20) %>%
    select(pathway) %>%
    unlist() %>%
    as.character()

View the selected pathways:

print(topPath)

Now subset fedupRes across all test sets for the pathways stored in topPath:

fedupPlot <- fedupRes %>%
    bind_rows(.id = "set") %>%
    separate(col = "set", into = c("set", "sign"), sep = "_") %>%
    subset(pathway %in% topPath) %>%
    mutate(pathway = gsub("\\%.*", "", pathway)) %>%
    mutate(sign = ifelse(status == "depleted", "none", sign)) %>%
    mutate(sign = factor(sign, levels = c("negative", "positive", "none"))) %>%
    group_by(set, pathway) %>%
    top_n(1, wt = fold_enrichment) %>%
    as.data.frame()

Plot via plotDotPlot, this time using set as our x-axis variable:

p <- plotDotPlot(
        df = fedupPlot,
        xVar = "set",
        yVar = "pathway",
        xLab = NULL,
        fillVar = "sign",
        fillLab = "Genetic interaction",
        fillCol = c("#6D90CA", "#F6EB13", "grey80"),
        sizeVar = "fold_enrichment",
        sizeLab = "Fold enrichment") +
    theme(
        panel.grid.major.y = element_blank(),
        axis.text.x = element_text(face = "italic", angle = 90,
        vjust = 0.5, hjust = 1))
print(p)

Ok cool, that's easier to look at than the plot before. Now let's summarize these pathways even more efficiently using EnrichmentMap!

Enrichment map

First, make sure to have Cytoscape downloaded and and open on your computer. You'll also need to install the EnrichmentMap (≥ v3.3.0) and AutoAnnotate apps.

Then format results for compatibility with EnrichmentMap using writeFemap:

resultsFolder <- tempdir()
writeFemap(fedupRes, resultsFolder)

Prepare a pathway annotation file (gmt format) from the pathway list you passed to runFedup using the writePathways function (you don't need to run this function if your pathway annotations are already in gmt format, but it doesn't hurt to make sure):

gmtFile <- tempfile("pathwaysGMT", fileext = ".gmt")
writePathways(pathwaysGMT, gmtFile)

Cytoscape is open right? If so, run these lines and let the plotFemap magic happen:

netFile <- tempfile("fedupEM_geneMulti", fileext = ".png")
plotFemap(
    gmtFile = gmtFile,
    resultsFolder = resultsFolder,
    qvalue = 0.05,
    chartData = "DATA_SET",
    hideNodeLabels = TRUE,
    netName = "fedupEM_geneMulti",
    netFile = netFile
)

To note here, the EM nodes were coloured manually (by a similar palette of colours passed to plotDotPlot) in Cytoscape via the Change Colors option in the EM panel. A feature for automated dataset colouring is set to be released in version 3.3.2 of EnrichmentMap.

This has effectively summarized the 244 pathways from our dot plot into 27 unique biological themes (including 7 unclustered pathways). We can now see clear themes in the data pertaining to negative and positive genetic interactions related to our genes of interest.