Systems Biology
RNA-seq data analysis
IV. Gene ontology,
Gene set enrichment analysis, &
Immune profiling
April 26, 2022
Bioinformatics & Genomics Lab
Department of Life Science
Hanyang University
Sang-Ho Yoon
Bioinformatics & Genomics, the BIG lab.
RNA-seq analytic pipeline
Bioinformatics & Genomics, the BIG lab.
RNA-seq analytic pipeline
FastQC
Sickle
STAR
FeatureCounts
DESeq2
T cell
B cell
T cell
B cell
Alignment of RNA-seq reads to the reference genome
Quantification of reads at gene level
Differential expression between conditions
Quality trimming
Quality assessment
Bioinformatics & Genomics, the BIG lab.
4
Marker genes identified using DEG analysis
- Investigating individual DEGs is always a good starting point for data interpretation
- However, there are hundreds of significant DEGs for each condition/cell type
🡪 How do you summarize the huge number of DEGs in your results?
Meaningful genes?
🡪 1067 genes
|log2FC| ≥ 2
&
adjusted P ≤ 0.05
🡪 1787 genes
Bioinformatics & Genomics, the BIG lab.
5
Gene Ontology (GO)
Gene Ontology consortium
- First published in 2000, Ashburner et al. proposed knowledge-based three ontologies
🡪 biological process, cellular component, molecular function
- The database is constantly being updated, and now covers various model organisms
- Molecular function (MF)
🡪 gene’s jobs, functions, or abilities
🡪 e.g., transporter, transcription factor, motifs, etc.
- Biological process (BP)
🡪 Signaling pathways or phenotypes
🡪 e.g., cell differentiation, cell death, development, etc.
- Cellular component (CC)
🡪 Localization of gene/protein in a cell
🡪 e.g., nucleus, cytosol, cell membrane, etc.
Bioinformatics & Genomics, the BIG lab.
6
Gene Ontology (GO)
Relationship b/w GO terms
- GO terms form hierarchical structures based on functions of gene/protein
Root
Leaves
- One can explore these relations using GO browsing tools (AmiGo)
Bioinformatics & Genomics, the BIG lab.
7
Gene Ontology (GO)
Relationship b/w GO terms
- DEGs are compared against gene sets (GO term) whether genes of interest are
overrepresented in a predefined gene set
Enrichment test
(Fisher’s exact test or Chi-squared test)
Bioinformatics & Genomics, the BIG lab.
8
Connect VPN and open your Shell
호스트: 166.104.118.161
포트 번호: 22
User: biguser
Password: biglab2428
1.
2.
3.
4.
P
RNA-seq data analysis: GSEA
and transfer <immune_cells_deseq2.txt> to your laptop using Xftp
(DESeq2 result file)
Bioinformatics & Genomics, the BIG lab.
9
P
RNA-seq data analysis: Gene Ontology (GO)
Open <immune_cells_deseq2.txt> from the last week’s result
Bioinformatics & Genomics, the BIG lab.
10
P
RNA-seq data analysis: Gene Ontology (GO)
T cell DEGs; select genes log2FoldChange >= 4 using the “Filter” function in Excel
Copy all gene symbols and paste into the Panther GO analysis tab
Biological process
Homo sapiens
Bioinformatics & Genomics, the BIG lab.
11
P
RNA-seq data analysis: Gene Ontology (GO)
Bioinformatics & Genomics, the BIG lab.
12
Gene Set Enrichment Analysis (GSEA)
- a computational method that determines whether an a priori defined set of genes
shows statistically significant, concordant differences between two biological states
Bioinformatics & Genomics, the BIG lab.
13
Gene Set Enrichment Analysis (GSEA)
- a computational method that determines whether an a priori defined set of genes
shows statistically significant, concordant differences between two biological states
Bioinformatics & Genomics, the BIG lab.
14
Gene Set Enrichment Analysis (GSEA)
- Molecular signature database (MSigDB) curates and distribute gene sets
Bioinformatics & Genomics, the BIG lab.
15
Gene Set Enrichment Analysis (GSEA)
- Gene sets are categorized into eight different clusters by different functionalities
- Hallmark gene sets summarize and represent
specific well-defined biological states or processes and display coherent expression.
Bioinformatics & Genomics, the BIG lab.
16
Gene Set Enrichment Analysis (GSEA)
- a computational method that determines whether an a priori defined set of genes
shows statistically significant, concordant differences between two biological states
Bioinformatics & Genomics, the BIG lab.
17
Gene Set Enrichment Analysis (GSEA)
Bioinformatics & Genomics, the BIG lab.
18
Gene Set Enrichment Analysis (GSEA)
1) Rank genes in the dataset using expression
The condition of interest
- First create a gradient or rank list of genes based on the degree of correlation to the phenotypes in data
- Genes on the left-hand side are highly expressed on the cancer group (phenotype) and lowly expressed on the normal group
Bioinformatics & Genomics, the BIG lab.
19
Gene Set Enrichment Analysis (GSEA)
2) Locate individual genes in a gene set in the ranked data
each vertical line indicate a gene
Bioinformatics & Genomics, the BIG lab.
20
Gene Set Enrichment Analysis (GSEA)
3) Calculate enrichment score (ES) for the gene set
Score goes up when you encounter a gene from the gene set.
Otherwise, score goes down
Bioinformatics & Genomics, the BIG lab.
21
Gene Set Enrichment Analysis (GSEA)
3) Calculate enrichment score (ES) for the gene set
- Kolmogorov-Smirnov (K-S) running sum statistic is computed to get ES
- ES = the greatest positive deviation of the running sum across all N genes
FC = Cancer/Normal
Bioinformatics & Genomics, the BIG lab.
22
Gene Set Enrichment Analysis (GSEA)
3) Calculate enrichment score (ES) for the gene set
Normalized ES (NES) = actual ES / mean(ESs against all permutations of the dataset)
Bioinformatics & Genomics, the BIG lab.
23
Gene Set Enrichment Analysis (GSEA)
4) Perform iterative enrichment analysis using permuted data to generate background
Permute data 100~1000 times;
The number of total permutation correlates statistical power
Bioinformatics & Genomics, the BIG lab.
24
Gene Set Enrichment Analysis (GSEA)
5) Compare the actual ES to the ES histogram generated by permuted data
🡪 this will give you a global P-value whether the gene set is enriched in the group
6) Normalize ES to compare different gene sets
Normalized ES (NES) = actual ES / mean(ESs against all permutations of the dataset)
(to account for differences in gene set size and correlations between gene sets and the expression dataset )
🡪 NES is affected by permutation method, the number of permutations, or the size of the expression dataset
Bioinformatics & Genomics, the BIG lab.
25
Connect VPN and open your Shell
호스트: 166.104.118.161
포트 번호: 22
User: biguser
Password: biglab2428
1.
2.
3.
4.
P
RNA-seq data analysis: GSEA
Bioinformatics & Genomics, the BIG lab.
26
Navigate into your working space
P
$ cd systems_biology_2022/YOUR_DIRECTORY
$ mkdir session_8
$ cd session_8
RNA-seq data analysis: GSEA
Copy these files and transfer to your laptop using Xftp
cp ~/systems_biology_2022/TA/session_8/TCGA_COAD_FPKM_5PTs.txt .
cp ~/systems_biology_2022/TA/session_8/TCGA_COAD_FPKM_5PTs.label.txt .
cp ~/systems_biology_2022/TA/session_8/gsea.label.cls.txt .
Bioinformatics & Genomics, the BIG lab.
27
FPKM expression of five colon cancer patients from The Cancer Genome Atlas database
P
RNA-seq data analysis: GSEA
TCGA_COAD_FPKM_5PTs.txt
gsea.label.cls.txt
🡪 A DESCRIPTION column is required for GSEA (you can leave this column empty)
A cls file is required for GSEA and specifies sample phenotype (label); 0 for Tumor, 1 for Normal
0 1 0 1... label of samples in corresponding order (tumor: 0, normal:1)
10: Total number of samples
2: number of labels (normal, tumor)
1: always 1. required
Bioinformatics & Genomics, the BIG lab.
28
P
RNA-seq data analysis: GSEA
Download GSEA app and the Hallmark gene set from
Downloads tab 🡪 register your e-mail to access
Bioinformatics & Genomics, the BIG lab.
29
P
RNA-seq data analysis: GSEA
Download a compatible version of GSEA
Bioinformatics & Genomics, the BIG lab.
30
P
RNA-seq data analysis: GSEA
Scroll down and find the hallmark gene sets in the MSigDB section
Download “h.all.v7.5.1.symbols.gmt”
Bioinformatics & Genomics, the BIG lab.
31
P
RNA-seq data analysis: GSEA
Install and open GSEA app
1. Click on “Load data”
Bioinformatics & Genomics, the BIG lab.
32
P
RNA-seq data analysis: GSEA
2. Upload your 1) expression matrix, 2) cls file, and 3) gmt file (gene sets)
3. Load these files!
Bioinformatics & Genomics, the BIG lab.
33
P
RNA-seq data analysis: GSEA
4. Go to “Run GSEA” tab
1
2
3
Bioinformatics & Genomics, the BIG lab.
34
P
RNA-seq data analysis: GSEA
4. Go to “Run GSEA” tab
Click on “Success”
Bioinformatics & Genomics, the BIG lab.
35
P
RNA-seq data analysis: GSEA
5. Browse the result
All results in single snapshot
0 = Tumor in the cls file; 20 gene sets are enriched in the tumor samples
# of significant gene sets
1 = Normal in the cls file; 30 gene sets are enriched in the normal samples
Results w/ statistics
Bioinformatics & Genomics, the BIG lab.
36
P
RNA-seq data analysis: GSEA
- Tumor-enriched 20 gene sets
Proliferation
Proliferation
Stress responses
Stress responses
Cellular development
Proliferation
Tumor metastasis
Metabolic pathway
Proliferation
Bioinformatics & Genomics, the BIG lab.
37
P
RNA-seq data analysis: GSEA
4. Go to “Run GSEA” tab
Click on “Show results folder”
Bioinformatics & Genomics, the BIG lab.
38
Immune profiling using CIBERSORT deconvolution
RNA-seq data is usually consisted of heterogeneous cell types
🡪 We call this data as “bulk” sequencing data
Tumor is a complex ecosystem comprised of malignant cells and non-malignant cells
🡪 Tumor microenvironment (TME)
Bulk RNA-seq data measures averaged gene expression of heterogeneous cells
Non-malignant cells influence metastasis,
drug resistance, and therapeutic responses
Hugo Gonzalez et al., Genes & development (2018)
🡪 How do you measure these cells in RNA-seq?
T. L. Whiteside, Oncogene (2008); Douglas Hanahan and Robert A. Weinberg, cell (2011)
Bioinformatics & Genomics, the BIG lab.
39
Immune profiling using CIBERSORT deconvolution
Computational dissection of bulk RNA-seq data
🡪 Marker gene expression, enrichment test of immune gene sets, or deconvolution
- Deconvolution is a technique isolating individual signals from a mixture
e.g., isolating flute soundtrack from orchestra
Bulk data
Reference
?
CIBERSORT
Aaron M. Newman et al., Nature methods (2015)
LM22: 22 immune cell types
Bioinformatics & Genomics, the BIG lab.
40
Immune profiling using CIBERSORT deconvolution
Infiltration of immune cells at pan-cancer level
- CIBERSORT analysis was performed for ~18000 samples of 39 human cancers
- Infiltration of plasma cells (good) and neutrophils (bad) as survival predictors
Andrew J. Gentles et al., Nature medicine (2015)
- LM22: the reference matrix of CIBERSORT, describes 22 immune cell types
Bioinformatics & Genomics, the BIG lab.
41
P
RNA-seq data analysis: CIBERSORT deconvolution
- CIBERSORT analysis of TCGA colon cancer 5 patients' data
- Register your email (Hanyang mail) in the CIBERSORT website
Bioinformatics & Genomics, the BIG lab.
42
P
RNA-seq data analysis: CIBERSORT deconvolution
- After registration, click on “Upload Files” in the Menu tab
- And upload your expression matrix (FPKM normalized)
The DESCRIPTION column should be deleted before uploading
Select “Mixture”
File name
Bioinformatics & Genomics, the BIG lab.
43
P
RNA-seq data analysis: CIBERSORT deconvolution
- Go to Run CIBERSORTx
Click on the tab #2
Bioinformatics & Genomics, the BIG lab.
44
P
RNA-seq data analysis: CIBERSORT deconvolution
- Specify files and parameters, then run CIBERSORT
Your expression matrix
LM22 as your reference matrix
100 permutations
Bioinformatics & Genomics, the BIG lab.
45
P
RNA-seq data analysis: CIBERSORT deconvolution
Bioinformatics & Genomics, the BIG lab.
46
P
RNA-seq data analysis: CIBERSORT deconvolution
Bioinformatics & Genomics, the BIG lab.
47
P
RNA-seq data analysis: CIBERSORT deconvolution
T #1
N #1
T #2
N #2
T #3
N #3
T #4
N #4
T #5
N #5
Macrophages T>>N
CD8 T cells T<<N
B cells T<<N
(purple to red)
(orange)
(yellow)
& Depletion of adaptive immunity (tumor killing) in tumors
- Accumulation of tumor-associated macrophages (tumor supporting)
Bioinformatics & Genomics, the BIG lab.
48
Homework
- No assignment in this week
- Good luck on your mid-term exam ☺