reduce redundancy of enriched GO terms

[GuangchuangYu]

To simplify the enriched result, we can use slim version of GO and use enricher() function to analyze.

Another strategy is to use GOSemSim to calculate similarity of GO terms and remove those highly similar terms by keeping one representative term.

The criteria of selecting representative term can be:

• most informative term (need pre-calculated IC data, only available for those internally supported organisms in GOSemSim; can be extended to un-supported organisms).
• most significant term (as in REVIGO)

I prefer using the second criteria for it's more intuitive and more easy to implement for those not internally supported by GOSemSim.

I propose to define a function to simplify the output from enrichGO by removing redundant GO terms.

simplify <- function(enrichResult, cutoff=0.7, by="p.adjust", select_fun=min) {
     ## GO terms that have semantic similarity higher than `cutoff` are treated as redundant terms
     ## select one representative term by applying `select_fun` to feature specifying by `by`.
     ## user can defined their own `select_fun` function.

     ## return an updated `enrichResult` object.
}

Any comment/suggestion is welcome.

Reference:

• https://www.biostars.org/p/47672/
• http://www.sciencedirect.com/science/article/pii/S0002929714002626

[GuangchuangYu]

finished in 7e32c7d

I will close the issue. Comment is still welcome.

[GuangchuangYu]

output from compareCluster is also supported, see983262b.

[GuangchuangYu]

see also the blog post

[fssdlyl001]

Hi, I applied simplify() function to compareCluster result, but return an error message: "Error in FUN(X[[i]], ...) : unused argument (organism = "human")". If I used enrichGO result as input, simplify worked. Could you help me solve this problem? Thank you.

[GuangchuangYu]

it was fixed in github version which will be released on Wednesday in BioC 3.4.

[fssdlyl001]

Thank you very much. Just one more question: which "semData" should I use for simpifying compareCluster results? If it needs to set manually, what's the default semData of simplify()?

[GuangchuangYu]

you need to refer to the GOSemSim vignette.

If you use measure = 'Wang', you don't need to pass semData.

[fssdlyl001]

But when I run simplify(x, measure="Wang"), it also return an message: Error in FUN(X[[i]], ...) : argument "semData" is missing, with no default. If I add 'semData = NULL', I got Error in match.arg(ont, c("BP", "CC", "MF")) : 'arg' must be of length 1. Where did I go wrong?

[GuangchuangYu]

data(gcSample)
x = compareCluster(gcSample, fun='enrichGO', ont='MF', OrgDb='org.Hs.eg.db')
y <- simplify(x, measure='Wang', semData=NULL) 

works.

pls follow the guide and provide a reproducible example.

[tigerxu]

Hi, I applied the simplify() function to reduce redundancy of the enriched GO terms produced by enrichGO. The codes below runs fine without error. But I noticed that several redundant GO terms were not filtered out. The example is below

genedata <- read.table("genedata.txt", quote=NULL, header=TRUE, check.names=F, sep="\t")
head(genedata)
eg = bitr(genedata$Protein IDs, fromType="UNIPROT", toType="ENTREZID", OrgDb="org.Hs.eg.db")
egobp <- enrichGO(gene = eg$ENTREZID,
OrgDb = org.Hs.eg.db,
ont = "BP",
pAdjustMethod = "BH",
pvalueCutoff = 0.01,
qvalueCutoff = 0.01,
readable = TRUE)
egobp2 <- simplify(egobp, cutoff=0.6, by="p.adjust", select_fun=min)
head(egobp2)

I show the first six GO terms below
Count
GO:0006335 24
GO:0034723 24
GO:0045861 64
GO:0051290 25
GO:0000183 24
GO:0002576 36

I compared the similarity of GO terms using GOSemSim and the result is as follows

goSim("GO:0006335", "GO:0034723", semData=hsGObp, measure="Jiang")
[1] 1
goSim("GO:0006335", "GO:0034723", semData=hsGObp, measure="Wang")
[1] 0.671

It looks that one of "GO:0006335" and "GO:0034723" should be filtered but actually not. Can you help to look at this problem? Any advice is very appreciated.

Thanks for developing clusterProfiler!

Zhuofei Xu

[GuangchuangYu]

pls follow https://guangchuangyu.github.io/2016/07/how-to-bug-author/ and provide a reproducible example.

[msquatrito]

Hi,
I've been trying to use the simplify function and I'had similar issue to @tigerxu. I've realised that in my case the problem was due to the fact that GO terms with the same "p.adjust" are not filtered out.
I've solved it using simplify(..., by = "pvalue").
Best,
Massimo

[tigerxu]

thanks for sharing this info, Massino! I will try again.
Zhuofei

[hmontenegro]

Doesn't simplify work with gseaResult objects?

cc.simp <- simplify( ego.cc, cutoff = 0.7, by = "p.adjust", select_fun = min )

Error in (function (classes, fdef, mtable) :
unable to find an inherited method for function ‘simplify’ for signature ‘"gseaResult"’

[haiyueliu]

Hi,

I am using simplify function, but it comes to errors.

ego2_52_slim <- simplify(ego2_52, cutoff=0.7, by="p.adjust", select_fun=min)
Error in .local(x, ...) :
simplify only applied to output from enrichGO...

I am sure the input is result from enrichGO. Could you please help me to have a look of this?

And also, you mentioned that you prefer using the second criteria which is "REVIGO", but here in simplify, you integrated GOSemSim, right?.

Thanks,
Haiyue

[Lucyyang1991]

It takes me really long time to run the simplify function, it that normal?

[sjidong12]

I have a similar problem with "Lucyyang1991". It takes quite long time to run the simplify function. Is that normal?

[davemcg]

simplify does not work with enrichGO output if the ontology used is "ALL"

[steffenheyne]

same here...also when I don't use "ALL" but like ont="BP", simplify never finishes for me

[VictorGoitea]

the same... ont="BP", simplify never finishes for me

[steffenheyne]

actually the problem seems to be that now (was there a change?) eg. enrichGO() returns an object that stores in ego@result also all unsignificant results. When I strip this down to the sig.-only results
by eg.:

ego@result = ego@result[ego@result$qvalue<=ego@qvalueCutoff,]

simplify finishes normally. But I don't know if the result of simplify is still correct with that change.
@GuangchuangYu ?

I just also strip down ego@geneSets = data.frame() as the result df contains all the genesets. With that I get a much smaller objects. But I think this is not used by simplify at all.

[GuangchuangYu]

@steffenheyne thanks for pointing this out. Yes, the result should be strip down before simplify. Will fix it.

[VictorGoitea]

Thanks Steffen! Your suggestion worked perfect.

[xxz19900]

simplify does not work with enrichGO output if the ontology used is "ALL"

Have you solved this problem? I also encountered such condition.

[steffenheyne]

@xxz19900 simplify only works within a sub-ontology like BP, MF or CC, not with ont= 'ALL', like

x <- enrichGO(dat, ont = 'BP', OrgDb = 'org.Hs.eg.db')
y <- simplify(x)

[kamalmdmostafa]

I am having the following error when trying to simplify Arabidopsis GO BP data:

egoBP_earlyresponse <- enrichGO(gene = earlyresponse,

•                             universe = allOE_genes,
  

•                             keyType = "TAIR",
  

•                             OrgDb = org.At.tair.db, 
  

•                             ont = "BP", 
  

•                             pAdjustMethod = "BH", 
  

•                             pvalueCutoff= 0.01,
  

•                             qvalueCutoff = 0.05,
  

•                             readable = FALSE)
  

cluster_summaryBP_earlyresponse <- data.frame(egoBP_earlyresponse)
#simplify

simplify(egoBP_earlyresponse, cutoff = 0.7, by = "p.adjust", select_fun=min)
Error in simplify(egoBP_earlyresponse, cutoff = 0.7, by = "p.adjust", :
unused arguments (cutoff = 0.7, by = "p.adjust", select_fun = min)

Any suggestion will be appreciated .

[laijen000]

Hi Dr. Yu (@GuangchuangYu ),

I would like to use GOSemSim to select the most informative term when comparing lists of GO IDs using mgoSim(). I am using the mouse genome and calculated IC scores: mmGO <- godata('org.Mm.eg.db', ont='BP').
Once we obtain the similarity matrix, is there a way to filter it to select just the most informative term above a certain similarity cutoff?

Because my GO IDs are not from using enrichGO, I am unable to use the simplify function. So I would like to try GOSemSim instead. Thank you!

[claraina]

I am having the same problem as @kamalmdmostafa, getting the error "unused arguments (cutoff = 0.7, by = "p.adjust", select_fun = min)" even though it seems to work with another set of genes. I don't think it's an issue with my enrich object (ego) as it is still showing the relevant GO Terms.

EEDDOC_ego <- enrichGO(gene = EEDDOC_gene.df$ENTREZID,
OrgDb = org.Mm.eg.db,
keyType = 'ENTREZID',
ont = "BP",
pAdjustMethod = "BH",
pvalueCutoff = 0.05,
qvalueCutoff = 0.05,
readable = TRUE)

sEEDDOC_ego <- simplify(EEDDOC_ego, cutoff=0.7, by="p.adjust", select_fun = min)

Error in simplify(EEDDOC_ego, cutoff = 0.7, by = "p.adjust", select_fun = min) :
unused arguments (cutoff = 0.7, by = "p.adjust", select_fun = min)

[kamalmdmostafa]

@claraina could you solve this issue. Seems like they gave up on this issue. Any other way to solve this? Creating a new issue maybe?

[huerqiang]

@kamalmdmostafa #278

[huerqiang]

@kamalmdmostafa #278

[nikofleischer]

Hi, I love the clusterprofiler package and regularly use the gseaGO and gseaKEGG functions to investigate my scRNAseq experiment results. However, using these there is always the problem of having many related terms enriched. Would it be possible to implement a function that only selects the most specific term that is enriched in the result for visualization in the dotplot() function? And/or extend the simplify function to work on gsea objects?

Best, Niko

[huerqiang]

@sparsepix You could use the showCategory = Your_selectd_terms parameter in dotplot to display your selected terms.

[nikofleischer]

@huerqiang thanks for the tip. Is there a way to easily compute the terms to select (based on them being either the most specific one in one subtree or selecting only the term with the lowest FDR within the subtree or sth similar?)

[TylerSagendorf]

Hi, I love the clusterprofiler package and regularly use the gseaGO and gseaKEGG functions to investigate my scRNAseq experiment results. However, using these there is always the problem of having many related terms enriched. Would it be possible to implement a function that only selects the most specific term that is enriched in the result for visualization in the dotplot() function? And/or extend the simplify function to work on gsea objects?

Best, Niko

@sparsepix Hi Niko, here is some code that selects the most general term: #372

If you instead compute last_GO_level = unlist(lapply(GO_levels, max)) and then use

x <- simplify(x, cutoff = 0.7, by = "last_GO_level", select_fun = max),

you will get the most specific enriched terms.

If you are not looking for the most specific terms and just want to deal with the redundancy issue, use the msigdbr package to obtain non-redundant gene sets/pathways. Then, supply them to clusterProfiler::GSEA or fgsea::fgsea to perform FGSEA.

[deevdevil88]

hi @TylerSagendorf
you mentioned in your post at the end that i can obtain non-redundant gene-sets/pathways directly from misgdbr? but i couldn't find a specific function which allows me to download non-redundant GO BP or GO CC genesets. Unless , the genesets on MSigDB are non redundant already ?

[TylerSagendorf]

hi @TylerSagendorf you mentioned in your post at the end that i can obtain non-redundant gene-sets/pathways directly from misgdbr? but i couldn't find a specific function which allows me to download non-redundant GO BP or GO CC genesets. Unless , the genesets on MSigDB are non redundant already ?

@deevdevil88 The gene sets/pathways provided by MSigDB on their website and through the msigdbr package are non-redundant. They explain how GO terms are filtered in the v7.0 release notes. I only know for sure that the same type of procedure is applied to Reactome pathways, but they claim that the entire database has had its redundancy decreased (doi: 10.1016/j.cels.2015.12.004).

[deevdevil88]

@TylerSagendorf awesome ! Thank you for confirming and also including the relevant release notes. I use MSigdb loads but this non-redundancy of GO terms wasn't obvious. Thanks again.

[TylerSagendorf]

@deevdevil88 No problem! Glad I could help.