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!

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()