Gene Set Enrichment Analysis after DESeq2

Question

[Sreeraj Thamban]

Hello Biostars, Can anyone tell me how to prepare input data set for GSEA after Differential Gene Expression Analysis by DESeq2? How will I rank the genes? Should I rank based on log2FC or Adjusted P value? Is there any way to generate a GSEA ready data directly from DESeq2?. I was using topGo for gene ontology enrichment analysis before and recently came across GSEA. Which one is better GO enrichment analysis or GSEA? Even after going through the papers I couldn't find a significant difference between above two.

Thank you

[enxxx23]

I like DESeq2. It would be great to have in the future something like ROAST/CAMERA/GSEA in DESeq2 too!

[Lada]

HI Sreeraj,

I don't know what is your model organism. For humans, mouse, drosophila and similar stuff, I guess it's easy because you can use online available databases and ensemble annotations. I participated in one online course about RNAseq data analysis on HUMAN data so I can share what I learned if it's helpful for you. It's just that I still didn't try that on my own data but here's what I know.

For GSEA - Initially you install these stuff in R:

install.packages("BiocManager") BiocManager::install(version = "3.16") BiocManager::install("DESeq2") BiocManager::install("clusterProfiler") BiocManager::install("org.Hs.eg.db") --> this is an organism-specific annotation package, this one is for humans but for instance, you can maybe find some others here: http://geneontology.org/ OR you can make your own dataset if you are working with nonmodel. I'm not an expert and I am still learning but its DOABLE so here you can see a similar question from my side, maybe it will help you: GSEA on nonmodel organisms

You do DESeq on your Dseq Data Set (dds) and once you get the results you can do this to remove NA.

dds_results_filtered <-dds_results[complete.cases(dds_results),]

I think you should use p-adjusted values in your filtering because that is representing SIGNIFICANT differences.

Then you can make a data set just for significantly upregulated genes like this:

upreg <- rownames(dds_results_filtered)[dds_results_filtered$pvalue < 0.05 & dds_results_filtered$log2FoldChange > 0]

Then you load your libraries:

library(clusterProfiler) library(org.Hs.eg.db)

Then you do GSEA:

gsea <- enrichGO(upreg, OrgDb = org.Hs.eg.db, keyType = "ENSEMBL", ont = "BP", universe = rownames(dds_results_filtered))

than you can make a simplified view

gsea <- simplify(gsea)

extract the data from gsea in nice table, first terms listed are the most significant

gsea_df <- as.data.frame(gsea)

additionally for excel you can try this

write.table(gsea_df, file = "gsea.tsv", sep = "\t")

and finally to see a nice dot plot for example for the top 13 categories:

dotplot(gsea, showCategory =13)

And then you can repeat for downregulated.

Hope this helps.

Lada

[CTLong]

Just a comment, this is not really a gene set enrichment analysis. Rather an over-representation test.

Answer 1

[Prakash]

Hi Sreeraj

Genes can be ranked based on fold change and P value and that can be used in GSEA package.

you can use this R code for this purpose.

x <- read.table("DE_genes.txt",sep = "\t",header = T)
head(x)
x$fcsign <- sign(x$log2.fold_change.)
x$logP=-log10(x$p_value)
x$metric= x$logP/x$fcsign
y<-x[,c("Gene", "metric")]
head(y)
write.table(y,file="DE_genes.rnk",quote=F,sep="\t",row.names=F)

[langya]

in this case, what parameter should we input into GSEA?

[t-jim]

How would you handle NA values?

[Barry Digby]

.....

filtered <- na.omit(y)

write.table(filtered ,file="DE_genes.rnk",quote=F,sep="\t",row.names=F)

[g.birch15]

I have used this code but am struggling to obtain a table where the gene names are appearing as names and not numbers, for some reason it keeps saving a table with the ranks but no gene names.

[Barry Digby]

Try row.names = TRUE.

Also, you want to use col.names = FALSE, as GSEA complains when 'Gene' and 'metric' are in the rnk. file.

Answer 2

[Michael Love]

Here's a link to an answer I wrote a few years ago for using the gene set testing package goseq following DESeq2:

https://support.bioconductor.org/p/64811/#64815

I'm not sure what kind of input GSEA takes. I also like the methods behind ROAST and CAMERA from the limma package, but I haven't yet worked on integrating with those methods. For those two, you would need to run a limma analysis upstream.

[langya]

Hey I noticed in the newest DESeq2 version, the default setting of fold change is not shrunken fold change, may I ask why? i thought the shrunken fold change gives you higher confidence.

[Kevin Blighe]

You can generate these via lfcShrink()

Answer 3

[Oliver]

Another option to gene ranking is to use the "stat"-output, that is generated by DESeq2, since that takes the logFold-change, as well as the standard error into account.

Check this video to see how to directly use the DESeq2 output for GSEA: Video tutorial