gwangjinkim
commented
6 years ago Dear Guangchuan,
Thank you for clusterProfiler! It is a great work!
To my issue above: I could help myself out with a dirty hack:
####################################################
# to make possible `simplify` on gseaResult object
# I just inserted following code into the script:
# And then, `simplify()` can be applied on `gseaResult` objects of GO GSEA analyses (`gseGO()`).
require(magrittr) # because of the %<>% operator which is a pipe %>% with reassigning back `<-`
setMethod("simplify", signature(x="gseaResult"),
function(x, cutoff=0.7, by="p.adjust", select_fun=min, measure="Wang", semData=NULL) {
if (!x@setType %in% c("BP", "MF", "CC"))
stop("simplify only applied to output from enrichGO...")
x@result %<>% simplify_internal(., cutoff, by, select_fun,
measure, x@setType, semData)
return(x)
}
)
# from:
# https://github.com/GuangchuangYu/clusterProfiler/blob/master/R/simplify.R
# I added `packagename::` in front of some function names, since this code is outside the package
# and does not "see" some of the packages/package functions imported
# to the package environment
# but basically this is the unchanged `simplify_internal` function definition.
simplify_internal <- function(res, cutoff=0.7, by="p.adjust", select_fun=min, measure="Rel", ontology, semData) {
if (missing(semData) || is.null(semData)) {
if (measure == "Wang") {
semData <- GOSemSim::godata(ont = ontology)
} else {
stop("godata should be provided for IC-based methods...")
}
} else {
if (ontology != semData@ont) {
msg <- paste("semData is for", semData@ont, "ontology, while enrichment result is for", ontology)
stop(msg)
}
}
sim <- GOSemSim::mgoSim(res$ID, res$ID,
semData = semData,
measure=measure,
combine=NULL)
## to satisfy codetools for calling gather
go1 <- go2 <- similarity <- NULL
sim.df <- as.data.frame(sim)
sim.df$go1 <- row.names(sim.df)
sim.df <- tidyr::gather(sim.df, go2, similarity, -go1)
sim.df <- sim.df[!is.na(sim.df$similarity),]
## feature 'by' is attached to 'go1'
sim.df <- merge(sim.df, res[, c("ID", by)], by.x="go1", by.y="ID")
sim.df$go2 <- as.character(sim.df$go2)
ID <- res$ID
GO_to_remove <- character()
for (i in seq_along(ID)) {
ii <- which(sim.df$go2 == ID[i] & sim.df$similarity > cutoff)
## if length(ii) == 1, then go1 == go2
if (length(ii) < 2)
next
sim_subset <- sim.df[ii,]
jj <- which(sim_subset[, by] == select_fun(sim_subset[, by]))
## sim.df <- sim.df[-ii[-jj]]
GO_to_remove <- c(GO_to_remove, sim_subset$go1[-jj]) %>% unique
}
res[!res$ID %in% GO_to_remove, ]
}
########################################After that, it is possible to call simplify(gseGO.result, cutoff = 0.7, by = "p.adjust", select_fun = min).
GuangchuangYu
nikofleischer
DarioS
currently,
simplify()is not realized for gseaResult objects (aftergseGO()). Is there are good theoretical reason not to allow simplification of this result? Else I would love to see it realized for gseaResult objects ...