In protViz/prora: ORA, sigORA and GSEA functionalities for proteomic data

library(tidyverse)

Outline

• Over-representation analysis (ORA)
• Signature over-representation analysis (sigORA)
• Gene set enrichment analysis (GSEA)
• Aim of the package
• Structure of the generated reports

ORA | Over-representation analysis

ORA Workflow

plotdat <- data.frame(
  x = c(0, 2, 2, 4, 6)*1.5,
  y = c(0, 2, -2, 0, -0),
  lb = c("Protein Intensity \n Matrix", "(Clustering)", "Threshold", "Differentially \nregulated Set", "ORA")
)

ggplot(plotdat) +
  theme_void() +
  xlim(-1, 10) +
  ylim(-3, 3) +
  geom_segment(aes(x = c(0, 0, 2, 2, 4)*1.5, y = c(0, 0, 2, -2, 0), xend = c(2, 2, 4, 4, 6)*1.5, yend = c(2, -2, 0, 0, 0)), colour = "grey", alpha = 0.3) +
  geom_label(aes(x = x, y = y, label = lb), size = 5, fill = "lightgrey")

• (Hierarchical Clustering)
• Apply simple threshold (e.g. $\log_2{\mathrm{FC}}>0.5$, $p < 0.01$, $\dots$)
• Perform ORA with background specification

ORA vs GSEA

set.seed(666)
dat <- data.frame(
  ID = paste0("P", floor(runif(20, 1000, 9999))),
  Score = runif(20, -3, 3)
) %>% 
  mutate(` ` = "", `  ` = case_when(
    Score > 1 ~ "Differential",
    Score < 1 ~ "Non-differential"
  )) %>% 
  distinct(ID, .keep_all = TRUE) %>% 
  arrange(desc(Score))

dat  %>% 
  knitr::kable(digits = 2, format = "html") %>% 
  kableExtra::kable_styling(bootstrap_options = "striped", full_width = TRUE, position = "float_right", font_size = 16) %>% 
  kableExtra::row_spec(which(dat$Score > 1), color = "red") %>% 
  kableExtra::collapse_rows(columns = 3:4, valign = "middle") %>% 
  kableExtra::add_footnote(label = "\n Background consists of full list.", notation = "none")

ORA

• List of IDs sorted by some score

• Apply simple threshold (e.g. Score > 1)

  $\rightarrow$ Differentially regulated set + background

• Perform Fisher's Exact Test

GSEA

• List of IDs sorted by some score
• Give full information to GSEA
• Compute enrichment scores

Fisher's Exact Test

For a given pathway we have

• Pathway size
• Number of proteins of this pathway in the differentially regulated subset
• Number of proteins of this pathway in the background

tab <- matrix(c(12, 3, 7, 24), nrow = 2, byrow = TRUE)
dimnames(tab) <- list("GO Term" = c("Contained", "Not Contained"),
                      "Differentially expressed" = c("    Yes    ", "    No    "))

cat("Pathway GO:0003091")
tab
cat("p-value:", round(fisher.test(tab)$p.value, 5))

Disadvantages of ORA

• Redundancy: No 1:1 relation between proteins and pathways

• Multiplicity: p-value adjustment (i.e. via FDR)

sigORA | Signature over-representation analysis

Gene Pair Signatures

• Introduced by @Foroushani2013
• Search background and pathways for protein pairs unique to a given pathway

knitr::include_graphics("Figures/Table1_n.png")

• Afterwards conduct ORA using self-defined GPS repository

Properties of sigORA

• Compares weighted number of observed pairs of proteins (unique to a single pathway) in the differentially regulated subset to the number one would expect to show up in the specified background

• Background GPS repository can change for each experiment

• Fairly conservative in terms of hits

GSEA | Gene Set Enrichment Analysis

GSEA WebGestaltR

• Introduced by @Subramanian15545

• Using the WebGestaltR package [@webgestpkg]

knitr::include_graphics("Figures/jobsummary.png")

• Compares the enrichment of particular protein set members towards the upper and lower end of the provided ranked protein list

• For instance ranked by p-value, FDR, fold change

GSEA WebGestaltR

Enrichment plots for all pathways

knitr::include_graphics("Figures/GO:0000077.png")

GSEA WebGestaltR

GO slim summary

knitr::include_graphics("Figures/goslim_summary_Cells..p75...cmvto.png")

GSEA WebGestaltR

Barchart with enriched pathways

knitr::include_graphics("Figures/wg_bar.svg")

Disadvantages of WebGestaltR

• Online tool (connection issues)
• Last update to WebGestalt2019 broke a lot of R code
• Again last update to version 0.4.0 broke a lot of R code

• Permutation test involved $\rightarrow$ time consuming

  $\rightarrow$ limited reproducibility since no control over RNG seed

fgczgseaora | Package

Implemented functionalities

• sigora ORA \& sigORA: runSIGORA()

• WebGestaltR ORA: runWebGestaltORA()

• WebGestaltR GSEA: runWebGestaltGSEA()

So why write wrappers and additional functionalities around the existing packages? * `sigORA`: Hard to set up GPS repository ID mapping missing (UniProt $\rightarrow$ ENTREZ) Check ID mapping efficiency * `WebGestaltR`: More convenient reporting

fgczgseaora | Report structure

Folder hierarchy

Folder -- Contrast

• Folder -- Method + Target

  • html -- report
  • Rmd -- source file
  • WebGestaltR -- reports from WebGestaltR

ORA, sigORA & GSEA

Example report

References

protViz/prora documentation built on Dec. 12, 2021, 12:32 a.m.