New feature: simplify to parent GO terms

[TylerSagendorf]

I propose to develop a new function that uses a combination of simplify and getGOLevel to reduce the redundancy of GO terms. The main idea is that if two terms have a high similarity, the more general term will be retained. This is likely to select the parent GO terms. The code that I use for this is provided below, though it will need to be modified slightly. I also recommend saving the list of GO terms at each level and including it as an object in clusterProfiler to speed up computation on the user's end.

# For each ontology, get a list of terms at each level from 1:14
# Note: This fails to get terms from level 15 and above for some reason.
# Levels are defined here: https://david.ncifcrf.gov/content.jsp?file=FAQs.html#9

onts <- c("BP", "CC", "MF")

# Create a list of 3 lists
ont_list <- lapply(onts, function(ont) {
  # For levels 1-14, get the GO terms
  lapply(1:14, function(level) {
    clusterProfiler:::getGOLevel(ont, level)
  })
})
# Set the names to easily access elements later
names(ont_list) <- onts 

# Recommended to save ont_list

# x is an object of class gseaResult, in this case
ont <- x@setType # Use x@setType for GSEA results and x@ontology for SEA results

# For each ID, get the levels where it appears.
# Also get the first (most general) level where it appears.
x %<>%
  mutate(
    # All levels
    GO_levels = sapply(ID, function(ID) {
      # Alternatively, use min(which(...)) to just get the most general level
      which(
        unlist(lapply(ont_list[[ont]], function(s) ID %in% s))
      )
    }),
    # Most general level (lowest value)
    first_GO_level = sapply(GO_levels, min)
  )

# If GO terms have a high similarity, select the more general one
x <- simplify(x, cutoff = 0.7, by = "first_GO_level", select_fun = min)

[WJH58]

@TylerSagendorf Hi maybe this is an unrelated issue to your question, but which version did you use? I am using the latest version and I found getGOLevel() function was removed from this package. I have problem converting go id to go level.

[TylerSagendorf]

@TylerSagendorf Hi maybe this is an unrelated issue to your question, but which version did you use? I am using the latest version and I found getGOLevel() function was removed from this package. I have problem converting go id to go level.

I am using clusterProfiler 4.0.0, but getGOLevel is not an exported function. That is why I use the triple colon to access it. Also keep in mind that each ID may have multiple levels, depending on their relationship to other terms in the DAG.

[Starahoush]

Ty so much for this, as it is a very useful function! There are just 2 adjustmensts I had to do:

• There is a missing 's' in "clusterProfiler:::getGOLevel(ont, level)" -> shouldn't it be "onts"?
• There is an extra ')' in the which( ) function

After that, it worked perfectly. Looking forward for a implementation in the package