BIOS6660 shRNAseq
Gene Set Enrichment
Analysis
Tzu L Phang PhD
Robert Stearman PhD
April 16, 2014
Stearman Assessment
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
Genome annotation used was mm9 (from 7/2007). There’s more recent annotation
mm10 (12/2011). Was the chr19.fa sequence file derived from mm9 or mm10?
It would be nice to show a mapping table that included:
Chr19 reads
100%
Reads mapped to exons
X%
Reads not mapping to exons
Y%
One report I think had the reads mapped to exons numbers but didn’t do anything with it.
Of course, what does RNA-seq reads not mapping to exons mean?
Four methods used to get overlap gene list. No one had a final table that summarized the
values and ranges of fold-change and adjusted p-values. (Some of the FC values were
inverse others so this needs to be consistent). No one considered a 3 out of 4 overlap
list. No one had a heatmap of the summarized expression values.
No one had a supervised cluster based on the overlap genes.
Not everyone had R session recorded for reproducible research.
Most had a final heatmap of the TOP 15 genes from just the edgeglm method rather than
the 14 gene overlap.
Several people didn’t get the message about the spliceAlignment argument needed to be
on to maximize the reads mapped (~60% vs 90%).
If QC reports run and graphs shown but not much in way of interpretation.
Some of the heatmap coloring schemes were hard to read and non-standard.
Not everyone included a workflow chart to show the analysis path.
Tzu Assessment
• PDF can not be “knitr”
• Try to give more description of what you
observe on plot result …
Socialism
GSEA
SuperStar
Problems with the
SuperStar approach
Case 1: No significant genes; because the
relevant biological differences are modest
relative to the noise inherent to the microarray
technology
Case 2: Too many significant genes; difficult to
interpret and ad hoc approach depends on
biologist’s area of expertise
Case 3: Single-gene analysis may miss important
effects on pathways which normally comprised
of sets of genes acting in concert
Case 4: Gene lists produced from different labs
seldom shown concordances.
Gene Set Enrichment Analysis
(GSEA)
Considers an a priori defined GeneSet (e.g.,
members of a metabolic pathway), and
determines where these members are
significantly over-represented or enriched
at the top (or bottom) of a list of markers
ranked by the degree of correlation with a
specific phenotype or class distinction
Genes
Samples
The rows represent the samples or
chips, and the columns represent
the genes
Highly expressed in diseased
Diseased
Normal


Genes on the left side are highly
expressed on the top half (indicated
by red color) and lowly expressed on
the bottom half (indicated by blue
color). The reverse is shown on the
right-most genes
Created a gradient or ranked list
corresponding to the degree of
correlation with the two phenotypes
Lowly expressed in diseased


This is depicted nicely by the graph on the bottom of the figure,
where the positive ranks on the left represent the correlation to the
Disease phenotype and the negative ranks on the right signify the
correlation to the Normal phenotype
The graph also generates a rank gradient that represents the order of
the most up-regulated genes for the Disease sample on the left-most,
and the most up-regulated genes for the Normal samples on the rightmost
Diseased
Normal



Now, let’s hide the heatmap and replace the middle
part of the figure with genes from a specific geneset,
say genes from the Glycolysis pathway.
Each vertical blue bars represents a gene from the
pathway, being mapped on the same location as the
whole dataset
Again, genes that are located on the left side are highly
expressed on the Disease samples, and the opposite is
true for the right-most genes


Now, we are ready to demonstrate the GSEA
algorithm.
The walk down algorithm basically scans the ranked
gene list L, and when a member of S is encountered,
an Enrichment Score (ES) is registered. This is
illustrated on the top part of the figure below; when the
ES started to build upon encountering more genes
from the GeneSet S.

The more S genes is found, the higher the ES

But, when no S genes were encountered for a long
walk down, as indicated on the middle section of the
middle plot, the ES will decrease accordingly. In
other words, a high ES relies intimately with the
clustering of S genes in close proximity. In this
example, we would conclude that the S genes have
high degree of correlation with the Disease
phenotype since most of the ES was gained from the
left portion of the plot
Advantages of GSEA
• Agnostic to the type of gene set and the source
of annotation
• Operates on any ordered gene list
• Does not require the choice of a gene selection
threshold or the explicit definition of a statistically
significant marker set
• Uses distribution-free, non-parametric,
permutation-based test procedures with
increased statistical power
• Incorporates the permutation of phenotype
labels thereby preserving the “biological”
correlation structure of the markers
• Takes into account multiple hypotheses testing
of multiple gene sets
References
Subramanian, A., Tamayo, P., Mootha, V. K., Mukherjee, S., Ebert, B. L.,
Gillette, M. A., Paulovich, A., Pomeroy, S. L., Golub, T. R., Lander, E. S. &
Mesirov, J. P. (2005) Proc. Natl. Acad. Sci. USA 102, 15545-15550. Gene
set enrichment analysis: A knowledge-based approach for interpreting
genome-wide expression profiles
GSEA Broad Institute (MIT)
http://www.broadinstitute.org/gsea/index.jsp
GSEA download
BioC: gage package
• BIOS6660_Share/Week12_13_shRNAseq
– Week12_13_shRNAseq_Day2.R
– Gage.pdf
– We will be using built in dataset.
• Direct Download
Now, a Demo
Mark’s data
• BIOS6660_Share/Week12_13_shRNAseq
– cep701_AllshRNA_readCounts.txt
– Jihye_shRNA_lib_ALL_new.txt
cep701_AllshRNA_readCounts.txt
Jihye_shRNA_lib_ALL_new.txt
Convert Symbol to Entrez