GSEA

Practise using GSEA

---

Objective: perform a GSEA preranked analysis on mesenchymal and immunoreactive ovarian cancer subtypes

Time estimated: 2 h; taken 1.5 h;

Date started: 2023-3-20 ; completed: 2023-3-20

---

Process

Issue: cannot open docker with GSEA Solution: download software- GSEA v4.3.2 Mac App on GSEA official website

• Download data:
  • mesenchymal vs immuno rank file using github URL
  • newest baderlab geneset: [Human_GOBP_AllPathways_no_GO_iea_March_02_2023_symbol.gmt](http://download.baderlab.org/EM_Genesets/March_02_2023/Human/symbol/Human_GOBP_AllPathways_no_GO_iea_March_02_2023_symbol.gmt) with no IEA
• Upload using Load data
• Run GSEAPreanked
  • Collapse: No_Collapse
  • Basic fields: Maxsize 200
  • rest of the parameters used default, as the defaulted permutation method is gene set permutation.

Issue: no Max genesize for selection Solution: Scroll right for basic fields, show

Conditions:

• mesenchymal vs immuno rank file
• genesets from the baderlab geneset collection containing GO biological process, no IEA and pathways.
• maximum geneset size of 200
  • Exclude huge pathways, increase specificity of result
• minimum geneset size of 15
• gene set permutation
  • We use a ranked list instead of phenotype mutation which is not optimized for RNA seq datasets.
  • To calculate the NES values for all S and permutation, and compare where the actual ES is in this distribution using FDR and p value

Questions:

1. Explain the reasons for using each of the above parameters.
   • We used the newest baderlab geneset because
     • It is updated on a monthly basis, therefore it contain more pathways then GSEA default, and more up to date.
     • GO biological process is included, not included IEA for more credible pathways instead of electronic annotations.
   • Maximum geneset size is set to 200
     • We want to exclude huge pathways and increase specificity of our result.
   • Minimum geneset size is set to 15
     • We want to exclude extremely small pathways which might that may just contain a few number of genes. . This approach ensures that the resulting pathways are not too specific.
   • We chose gene set permutation
     • The other option: phenotype mutation is not optimized for RNA seq datasets
     • We used a ranked gene list, and we want to maintain the rank during the permutation of GSEA analysis
2. • What is the top gene set returned for the Mesenchymal sub type?
     • HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION%MSIGDBHALLMARK%HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION
     1. What is its pvalue, ES, NES and FDR associated with it.
        • pvalue: 0.000
        • ES: 0.87
        • NES: 2.59
        • FDR: 0.000
     2. How many genes in its leading edge?
        • 147 genes in its leading edge.
     3. What is the top gene associated with this geneset.
        • Top gene is FBN1.
   • What is the top gene set returned for the Immunoreactive subtype?
     • HALLMARK_INTERFERON_ALPHA_RESPONSE%MSIGDBHALLMARK%HALLMARK_INTERFERON_ALPHA_RESPONSE
     1. What is its pvalue, ES, NES and FDR associated with it.
        • pvalue: 0.000
        • ES: -0.86
        • NES: -2.90
        • FDR: 0.000
     2. How many genes in its leading edge?
        • 79 genes in its leading edge.
     3. What is the top gene associated with this geneset.
        • Top gene is PROCR.

Summary: GSEA account for all signals instead of the top differentiated ones, the negative and positive values account for different ovarian cancer subtypes and will be detected as phenotypes automatically.

References:

Mootha, V. K., Lindgren, C. M., Eriksson, K. F., Subramanian, A., Sihag, S., Lehar, J., Puigserver, P., Carlsson, E., Ridderstrale, M., Laurila, E., et al.  (2003). PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet 34, 267-273.