In jokergoo/simplifyGO: Simplify Functional Enrichment Results

library(knitr)
knitr::opts_chunk$set(
    error = FALSE,
    tidy  = FALSE,
    message = FALSE,
    warning = FALSE,
    fig.width = 5,
    fig.height = 5,
    fig.align = "center"
)
options("width" = 100)

The lt object contains a list of GO enrichment tables.

library(cola)
data(golub_cola) 
res = golub_cola["ATC:skmeans"]

library(hu6800.db)
x = hu6800ENTREZID
mapped_probes = mappedkeys(x)
id_mapping = unlist(as.list(x[mapped_probes]))

lt = functional_enrichment(res, k = 3, id_mapping = id_mapping)

names(lt)
head(lt[[1]][, 1:7])

We first demonstrate the new plot on the single enrichment table. To use the simplifyEnrichment package, we extract significant GO terms, and then call simplifyGO().

library(simplifyEnrichment)
df = lt[[1]]
go_id = df$ID[df$p.adjust < 0.01]
simplifyGO(go_id)

The plot looks good, but it still contains too many graphic contents. For example, the GO similarity heatmap is useful, but it takes too much space on the final plot. Here I developed a new function summarizeGO() which simplifies the enrichment results even more. The idea is that since we already have the GO clusters, with a certain statistic of enrichment, we can simply use its average for the GO cluster.

In the following example, we use -log10(p.adjust) as an enrichment measure. The heights of bars correspond to the mean of -log10(p.adjust) of GO terms in different GO clusters. On the left side, we still use the word clouds which efficiently show the general functions in each GO cluster.

l = df$p.adjust < 0.01
summarizeGO(df$ID[l], -log10(df$p.adjust)[l], axis_label = "average -log10(p.adjust)")

GO IDs can be attached to the numeric value vector, but this time, the value argument should be explicitely specified when calling summarizeGO().

v = -log10(df$p.adjust)
names(v) = df$ID
summarizeGO(value = v[l], axis_label = "average -log10(p.adjust)")

Beside -log10(p.adjust), we also suggest to use log2 fold enrichment as the enrichment measure. It is calculated as

$$ \log_2 \left( \frac{k/m_1}{m_2/n} \right) $$

where $k$ is the number of DE genes (if the genes of interest are DE genes) in a gene set, $m_1$ is the size of DE genes, $m_2$ is the size of the gene set, $n$ is the total number of genes in the universal set. Of course, the definition of $m_1$ and $m_2$ can be switched.

k = as.numeric(gsub("/\\d+$", "", df$GeneRatio))
m1 = as.numeric(gsub("^\\d+/", "", df$GeneRatio))
m2 = as.numeric(gsub("/\\d+$", "", df$BgRatio))
n = as.numeric(gsub("^\\d+/", "", df$BgRatio))
log2_fold_enrichment = log2(k*n/m1/m2)

summarizeGO(df$ID[l], log2_fold_enrichment[l], axis_label = "average log2(fold_enrichment)")

Of course, you can construct a named log2_fold_enrichment vector which only contains significant GO terms.

names(log2_fold_enrichment) = df$ID
summarizeGO(value = log2_fold_enrichment[l], axis_label = "average log2(fold_enrichment)")

For multiple GO enrichment results, simplifyGOFromMultipleLists() can be used to visualize and compare GO clusters.

simplifyGOFromMultipleLists(lt, padj_cutoff = 0.001)

summarizeGO() can also be used to simplify such plot. Now the value of value is a list of numeric named vectors which contains significant GO terms in each enrichment table:

value = lapply(lt, function(df) {
    v = -log10(df$p.adjust)
    names(v) = df$ID
    v[df$p.adjust < 0.001]
})
summarizeGO(value = value, axis_label = "average -log10(p.adjust)", legend_title = "-log10(p.adjust)")