Gene Ontologies (GO) are often used to guide the interpretation of high-throughput omics experiments, with lists of differentially regulated genes being summarized into sets of genes with a common functional representation. Due to the hierachical nature of Gene Ontologies, the resulting lists of enriched sets are usually redundant and difficult to interpret.
rrvgo
aims at simplifying the redundance of GO sets by grouping similar terms
based on their semantic similarity. It also provides some plots to help with
interpreting the summarized terms.
This software is heavily influenced by REVIGO. It mimics
a good part of its core functionality, and even some of the outputs are similar.
Without aims to compete, rrvgo
tries to offer a programatic interface using
available annotation databases and semantic similarity methods implemented in the
Bioconductor project.
Starting with a list of genes of interest (eg. coming from a differential expression analysis), apply any method for the identification of eneriched GO terms (see GOStats or GSEA).
rrvgo
does not care about genes, but GO terms. The input is a vector of enriched
GO terms, along with (recommended, but not mandatory) a vector of scores. If scores
are not provided, rrvgo
takes the GO term (set) size as a score, thus favoring
broader terms.
First step is to get the similarity matrix between terms. The function calculateSimMatrix
takes a list of GO terms for which the semantic simlarity is to be calculated,
an OrgDb
object for an organism, the ontology of interest and the method to
calculate the similarity scores.
library(rrvgo)
go_analysis <- read.delim(system.file("extdata/example.txt", package="rrvgo"))
simMatrix <- calculateSimMatrix(go_analysis$ID,
orgdb="org.Hs.eg.db",
ont="BP",
method="Rel")
The semdata
parameter (see ?calculateSimMatrix
) is not mandatory as it is
calculated on demand. If the function needs to run several times with the same
organism, it’s advisable to save the GOSemSim::godata(orgdb, ont=ont)
object,
in order to reuse it between calls and speedup the calculation of the similarity
matrix.
From the similarity matrix one can group terms based on similarity. rrvgo
provides the reduceSimMatrix
function for that. It takes as arguments i) the
similarity matrix, ii) an optional named vector of scores associated to each
GO term, iii) a similarity threshold used for grouping terms, and iv) an orgdb
object.
scores <- setNames(-log10(go_analysis$qvalue), go_analysis$ID)
reducedTerms <- reduceSimMatrix(simMatrix,
scores,
threshold=0.7,
orgdb="org.Hs.eg.db")
reduceSimMatrix
selects as the group representative the term with the higher
score within the group. In case the vector of scores is not available,
reduceSimMatrix
can either use the uniqueness of a term (default), or the
GO term size. In this case, rrvgo
will fetch the GO term size from the
OrgDb
object and use it as the score, thus favoring broader terms. Please
note that scores are interpreted in the direction that higher are better,
therefore if you use p-values as scores, minus log-transform them before.
Higher thresholds force higher similarity between terms of a groups, resulting in more groups containing less similar terms.
rrvgo
provides several methods for plotting and interpreting the results.
Plot similarity matrix as a heatmap, with clustering of columns of rows turned on by default (thus arranging together similar terms).
heatmapPlot(simMatrix,
reducedTerms,
annotateParent=TRUE,
annotationLabel="parentTerm",
fontsize=6)
The function internally uses pheatmap
,
and further parameters can be passed to this function.
Plot GO terms as scattered points. Distances between points represent the similarity between terms, and axes are the first 2 components of applying a PCoA to the (di)similarity matrix. Size of the point represents the provided scores or, in its absence, the number of genes the GO term contains.
scatterPlot(simMatrix, reducedTerms)