visualizing GSEA results with 'dotplot'

guidohooiveld commented 7 years ago

Hi Guangchuang, Maybe a somewhat naive question (feature request?), but is it somehow possible to visualize the results of a GSE run (of a single but also multiple runs) in a compareCluster-dotplot-like figure? I am asking because it would be really cool and helpful if one could represent the top up- and down-regulated genesets, ideally from multiple runs, in a single graph. For example, like you implemented for the compareCluster() function. I was triggered by this idea after reading your online vignette, specifically section 13.2 here, and knowing of the function merge_result() here.

I have something in mind like this picture (from the link to section 13.2 above): cc_updown ... which, 'translated' to GSE results, should rather show the up- and down-regulated gene sets form GSE analysis A and B (sorted either by significance or NES), and color coding representing significance, and node size equaling Gene Set size. (??)

Thanks for considering, Guido

As expected, the function dotplot() doesn't work with the output of a gene set enrichment analysis performed with DOSE.

> library(DOSE)
> data(geneList)
> x <- gseDO(geneList)
preparing geneSet collections...
GSEA analysis...
leading edge analysis...
done...

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

crazyhottommy commented 7 years ago

Hi, I had the same question. you can just plot it using ggplot2.

library(ggplot2)
library(dplyr)
library(stringr)

## count the gene number
gene_count<- x@result %>% group_by(ID) %>% summarise(count = sum(str_count(core_enrichment, "/")) + 1)

## merge with the original dataframe
dot_df<- left_join(x@result, gene_count, by = "ID") %>% mutate(GeneRatio = count/setSize)

## plot
library(forcats) ## for reordering the factor
ggplot(dot_df, aes(x = GeneRatio, y = fct_reorder(Description, GeneRatio))) + 
               geom_point(aes(size = GeneRatio, color = p.adjust)) +
               theme_bw(base_size = 14) +
        scale_colour_gradient(limits=c(0, 0.10), low="red") +
        ylab(NULL) +
        ggtitle("GO pathway enrichment")

Hope it helps!

Tommy

GuangchuangYu commented 7 years ago

thanks @crazyhottommy.

dot_df = dot_df[1:50,] ## small dataset
dot_df$type = "upregulated"
dot_df$type[dot_df$NES < 0] = "downregulated"

## from Tommy's code
p <- ggplot(dot_df, aes(x = GeneRatio, y = fct_reorder(Description, GeneRatio))) + 
               geom_point(aes(size = GeneRatio, color = p.adjust)) +
               theme_bw(base_size = 14) +
        scale_colour_gradient(limits=c(0, 0.10), low="red") +
        ylab(NULL) +
        ggtitle("GO pathway enrichment")

p + facet_grid(.~type)

screen shot 2016-12-21 at 5 26 03 pm

GuangchuangYu commented 7 years ago

I will add a dotplot method for GSEA result.

Any idea to improve?

guidohooiveld commented 7 years ago

Thanks to both of you! @crazyhottommy and @GuangchuangYu.

I played a bit with the code above to visualize the results of 2 GSE analysis. Since I am not really a programmer, the code is somewhat clumsy...

Although the code works, I would appreciate some refinement:

there are to many gene sets represented; in fact all gene sets that are significant are plotted [because of gseDO(), default settings: pvalueCutoff = 0.05,pAdjustMethod = "BH"]. This makes the graph unreadable. How to best select the (let's say) top 15 significant gene sets of both GSE results?
is there a better way to merge the results of both GSE runs? Thus not by using merge_result?

dose_merged_gsea_up_down

Code:
library(DOSE)
library(ggplot2)
library(dplyr)
library(stringr)
library(forcats)

data(geneList)
x.1 <- gseDO(geneList)
x.2 <- gseDO(geneList, by="DOSE", nPerm=1000) 

## count the gene number for both results
gene_count.x1 <- x.1@result %>% group_by(ID) %>% summarise(count = sum(str_count(core_enrichment, "/")) + 1)
gene_count.x2 <- x.2@result %>% group_by(ID) %>% summarise(count = sum(str_count(core_enrichment, "/")) + 1)

## merge with the original dataframes
dot_df.x1<- left_join(x.1@result, gene_count.x1, by = "ID") %>% mutate(GeneRatio = count/setSize)
dot_df.x2<- left_join(x.2@result, gene_count.x2, by = "ID") %>% mutate(GeneRatio = count/setSize)

## merge the two results
library(clusterProfiler)
merged.res <- as.data.frame(merge_result(list(fgsea=dot_df.x1, dose=dot_df.x2)))

## merged.res <- rbind(dot_df.x1, dot_df.x2) #This merging works but it does **not** include source of results (i.e. 'fgsea' or 'dose')

## Set up/downregulation
merged.res$type = "upregulated"
merged.res$type[merged.res$NES < 0] = "downregulated"

p <- ggplot(merged.res, aes(x = GeneRatio, y = fct_reorder(Description, GeneRatio))) + 
               geom_point(aes(size = GeneRatio, color = p.adjust)) +
               theme_bw(base_size = 14) +
        scale_colour_gradient(limits=c(0, 0.10), low="red") +
        ylab(NULL) +
        ggtitle("Disease Ontology enrichment")

p + facet_grid(.~Cluster+type) #Cluster and type are columns to 'split' on

ggsave("merged_GSE.png")

GuangchuangYu commented 7 years ago

For first issue, of course the showCategory parameter will work for it. For second issue, it should goes to clusterProfiler.

guidohooiveld commented 7 years ago

Thanks for your continuous feedback! However, I got lost by your 2 comments... Why: 1st issue: showCategory parameter is indeed utilized with the various plotting functions, including dotplot. However, dotplot doesn't accept gseaResult object yet.... I assume you mean it will work after you updated thedotplot function?? (https://github.com/GuangchuangYu/DOSE/issues/20#issuecomment-268476805).

2nd issue: using formula interface (as described here). Isn't it correct that you can only use the formula/grouping interface for over-representation (enrichment) analyses? i.e.: which pathways/ontologies are enriched in 'group' or 'othergroup'. But how to apply formula interface for GSE analysis, that uses a full, ranked dataset? In other words, how to use formula interface if you have e.g. 2 ranked lists of genes? Sorry if I am missing something obvious....

FYI: for now I got 1st issue 'working' by manual ordering and selecting on the merged dataframes (dot_df.x1 and dot_df.x2) before merging.

GuangchuangYu commented 7 years ago

issue 1 solved.

2016-12-22-233207_1280x800_scrot

Now dotplot supports gseaResult and showCategory and other parameters we familiar with dotplot method for enrichResult are all work also for gseaResult.

You can also pass the split parameter which will apply the showCateogry by spliting the results using specific parameter. Here .sign is reserved for the sign of NES (activated for >0 and suppressed for <0). So in this example, we plot 30 activated and 30 suppressed enriched terms.

For issue 2, I will elaborate more details when I have time to work it out.

crazyhottommy commented 7 years ago

nice work!

guidohooiveld commented 7 years ago

Indeed, very nice! Thanks Guangchuang!

alnf commented 7 years ago

I've been also trying to merge the results of two different GSE runs. I did smth very similar to what @guidohooiveld did, but I just copy pasted and slightly changed DOSE fortify function. Then I realized I can change order and by arguments, which are normally hidden, since dotplot calls fortify internally. I think accessing these arguments might be useful for both issues, i.e. dotplot and clusterProfiler.

Here is what I got using my own data: order=TRUE, by="Count" LCHP_gseGO_BP_count.pdf order=TRUE, by="GeneRatio" LCHP_gseGO_BP_gr.pdf order=FALSE LCHP_gseGO_BP.pdf

Update. If I correctly understood the source code by argument is implemented in clusterProfiler version of dotplot, but not in DOSE dotplot. Was there a specific reason for that?