Pathway Analysis using the Gene Set Enrichment Analysis (GSEA) Tool

Gene Set Enrichment Analysis is one of many approaches to the analysis of gene expression profile data and is described in a paper from workers at the Broad Institute.

The basic concept was prompted by the observation that studying individual genes showing the most significant difference in expression level between two states or phenotypes is lacking in mechanistic insight. Instead, it makes more sense to take a set of genes sharing some biological link, and ask the question – does the whole set show any statistically significant enrichment in those genes that have differential expression?

A gene set can be chosen, a priori, for a number of reasons e.g. the set of genes known to be influenced by over- or under-expression of a micro-RNA, or perhaps a set chosen based on chromosomal location, or genes for which molecular function, cellular component and / or biological process have been assigned using the controlled vocabularies of the Gene Ontology.

One advantage to the GSEA approach is that it is possible to incorporate your complete data set, not just those transcripts with an arbitrarily chosen differential expression threshold. I am sure that many people reading this will be thinking – “How can it be OK to use the complete dataset? Normally I would only consider genes with >2 (OR other favourite value)-fold differential expression.” The reason the approach is valid is that genes expressed at low levels or with large variance between replicates do not contribute to the main metric used by GSEA, the ‘enrichment score’ (ES).

GSEA works by first ranking the expression value for each gene by signal to noise ratio – calculating the difference between the mean values for samples representing each phenotype and scaling them by the sum of the standard deviations. This means that genes with large differences in expression level between different states and little variation between biological replicates are ranked highly.

The next step is that the ES, the primary statistic generated by GSEA, is calculated for each gene set – in the GSEA manual, which documents the software excellently, it states:

“All genes are first ranked by their signal to noise ratio, then the ES is calculated by “walking” down the ranked list of genes increasing a running-sum statistic when a gene is in the gene set and decreasing it when it is not. The magnitude of the increment depends on the correlation of the gene with a phenotype. The ES is the maximum deviation from zero encountered in walking the list. A positive ES indicates gene set enrichment at the top of the ranked list; a negative ES indicates gene set enrichment at the bottom of the ranked list.”

The ES values are normalised based on gene set size and then a false discovery rate is calculated, to give an estimated probability of false positives. GSEA uses a very relaxed default value of 25%, which is suitable for hypothesis generation with a relatively large number of biological replicates.

Scientists working on data from non-human samples can still use GSEA, but need to beware – the gene symbols used by GSEA are “translated” from their human equivalents i.e. identifiers used for genes from your species of interest represented on the microarray are converted into symbols for their human orthologues, then used in the analysis. Subramanian and colleagues claim that this conversion has little or no effect on the utility of GSEA; it has been used successfully in multiple non-human species, but of course this must be kept in mind when investigating results in detail.

For an excellent, in-depth, review of pathway tools, consult:

Khatri, P., Sirota, M., & Butte, A. J. (2012). Ten Years of Pathway Analysis: Current Approaches and Outstanding Challenges. PLoS Computational Biology, 8(2), e1002375. doi:10.1371/journal.pcbi.1002375

Another good source of advice on pathway analysis, especially for those familiar with the R statistics package is here.

Further reading

Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102:15545-15550

Xie X, Lu J, Kulbokas EJ, Golub TR, Mootha V, Lindblad-Toh K, Lander ES, Kellis M (2005) Systematic discovery of regulatory motifs in human promoters and 3[prime] UTRs by comparison of several mammals. Nature 434:338-345

Posted in Pathway analysis | 1 Response

Joys of editing academic science books

Image courtesy of ningmilo /

Or: “A beginner’s guide to herding cats”.

Consider this scenario: you are an academic scientist, in a busy research institute and your boss is invited to edit a book, but declines due to pressure of work; then suggests that it would look good on your CV. You agree, it would look good on your CV, so you commit yourself to editing your first multi-author academic science book.

So why is that a problem?

Getting authors on board

You want the best people to write the chapters. You Google some big-name experts and invite them to contribute a chapter to your book. They almost all decline, or fail to reply to your email. But, somewhat to your amazement, one agrees. However, this paragon of science then never, ever replies to any future contacts. So, you lower your sights and aim for good scientists, but not Nobel Prize winners. Finally, you get enough authors together to write the chapters around the topic that the publishers have given you – phew!

Getting authors to agree a deadline

Assuming it’s not unreasonable, everyone is usually relaxed about the deadline set. However, the real challenge is:

Getting them to meet the deadline

  1. This should be easy, right? Scientists are grown-up, professional people. Aren’t they? Well, sort of. In reality, academics typically over-commit themselves, doing not only research and teaching, but also writing grant funding applications, papers, reviews, book chapters, etc, etc. After all, the scientific mission statement is “publish OR be damned.”
  2. As the deadlines go past – “wooshh”, like passing cars, half your authors have submitted their chapters, the rest not. Now another sticky moment arrives – these are meant to be cutting edge reviews. State-of-the-Art. But this delay now means that the ‘good’ authors work is rapidly reaching its sell-by date. You may have to go crawling back to them to ask for updates. Which they are usually not too unhappy about, but you hate the loss of face.
  3. One more thing that I forgot to mention; as the editor, you have to READ these chapters. Worse still, you are expected to produce cogent critiques – what the author needs to add, remove, expand or contract. Even if the topic is on the fringe of your main expertise.

What happens if authors go AWOL?

What do you do when one of your authors decides that they are NOT going to write their chapter? Not simply procrastinate, fail to meet deadlines, but stop all communication. Disappear off the map. So, now you’re stuck – find another author(s) – more delay – write the chapter yourself? – but it’s too far outside your own area of expertise. So, eventually, you find someone else. Which means yet more delay.

Writing your own chapter

Oh, yes, you forgot that you agreed to write one of the chapters yourself. Oops. Oh well, not a problem. Offer co-authorship to one of your PhD students – they’ll be falling over themselves to get another publication on their CV. Or maybe not: no, they are not interested after all; obviously suspecting (correctly) that your aim is to let them write the whole thing, then submit the chapter to you for a little light editorial polishing.

Pleading with the publishers for more time

  1. You now hold the dubious record for the longest gestation period of a multi-author academic book in human history, excluding the Bible.
  2. ‘Please, sir, I want some more.’
  3. The publishers are not impressed, but quietly resigned, telling you to go away and come back when you meet a new deadline.

Losing your marbles and giving up completely

It’s all taking SO LONG – too few authors have submitted first drafts of their chapters. You start to get desperate – the original deadline was so long ago that you’ve forgotten it – the “new” deadline is also now history. You consider giving the whole thing up – apologise to the authors and the publishers and say the book can’t be finished. But your co-editor and the authors who have delivered on time are indignant – naturally enough they don’t want to see their work wasted – and insist that you go back to the recalcitrant scientists with a big stick. How do you threaten authors with a stick by email? Or by phone? However, a combination of the metaphorical big stick, pleas for mercy and piling on the guilt eventually work and all the chapters are delivered! Hooray.


So, now, you’re on the last lap. Or the last dregs – the soul-destroying process of assembling the index and proofreading. Once, a sub-editor with a scientific background might have written an index, but not now. Academic publishers want their pound of flesh, so this task is delegated to authors and editors. Authors select keywords from their chapters, with varying degrees of enthusiasm or accuracy, then the editor attempts to assemble them into something useful to the reader. Finally, a draft proof arrives by email. You are now heartily sick of every word, but a final spurt of enthusiasm drives you on and the book is finished.

One more thing – did I forget? – you don’t get paid – but you are given a few free copies of your own book. Such fun!


Posted in Light relief | 1 Response

How does a mutation in a transcription factor cause glue ear?

Acute otitis media, sometimes known as “glue ear”, is the most common bacterial infection in children and by 1 year of age about 60% of children will have had one episode. In some cases, children develop a chronic condition, which, despite the infection being cured, the “glue” doesn’t go away and causes deafness.  In an inherited mouse model of chronic glue ear the causative mutation has been shown to be in a gene encoding a transcription factor, Evi1.

The EVI1 protein has multiple domains, can repress or enhance expression of target genes and interact with many other proteins. Indeed, the multiplicity of known and potential interactions is a challenge to determining the role of the mutation.  There were clues, however, as to how this mutation might lead to disease from differences in phenotype e.g. mutant mice raised in a “clean” SPF animal facility were less likely to become deaf than those kept in the older, “dirty” animal house.

Did this mean that gene-environment interactions e.g. between immune system and microbes, influence disease susceptibility? It was also known that mutant mice showed high levels of influx of neutrophils into their middle ear cavities (inflammation), but it was unclear whether EVI1 was acting directly or indirectly in this process. Possible answers to these questions came recently from studies in cultured cells, showing that EVI1 can act as an inhibitor of one of the key proteins regulating inflammation, another transcription factor, nuclear factor kappa B (NFkB).  EVI1 binds to to one of the subunits of NFkB and interferes with a critical protein modification, acetylation.  However, EVI1 does not acetylate proteins directly, so other factors must be involved. What were those other factors?

I combined public and unpublished data using literature searches and open source software e.g. iRefWeb in order to identify steps in the NFkB signalling pathways that might be disturbed by the mutation in EVI1.  The novel target proteins and starting points for drug development I discovered are suitable for testing in this preclinical model of chronic otitis media.

Read our testimonial from Dr Michael Cheeseman.


Posted in Pathway analysis, Target discovery | Leave a comment

Target discovery in childhood-onset asthma

Asthma is caused by a combination of environmental and genetic influences, but the specific factors are poorly understood. A significant “hit” detected in a genome-wide association scan (GWAS) for childhood asthma led a client to believe that one gene might be partially responsible. Proving that this genetic association really was causing asthma was, however, difficult. Firstly, no one knew the function of the protein made by the gene and secondly, changing genes in humans to test a hypothesis, rather than as therapy, is technically challenging & ethically questionable, especially in children. Fortunately, mice share about 90% of their genes with humans, so scientists “knocked-out” the equivalent gene, then tested whether these animals behaved like children with asthma. The short answer is – they didn’t. In lung-function tests that would have had asthmatics reaching for their inhalers, the knock-out mice were completely normal. So, what was going on? Were mice not enough like humans? Was this the wrong gene?

For this project, I went back to first principles – what was the evidence supporting the idea that this gene was responsible for increased asthma risk? Digging through the online literature, in particular papers from other groups studying the same gene and supplementary material not available in print, there were suggestions that the genetic effects were more complex. I found evidence that two other genes nearby were either more or less transcriptionally active in asthmatics and so might play a role in susceptibility to asthma. Furthermore, using data from the ENCODE project, I found that the regulatory element predicted to control these genes was conserved in mice, so it would be possible to test the predictions experimentally.

This suggested a novel therapeutic target – altering the activity of a cluster of genes, rather than just one, might alter disease risk.


Posted in Target discovery | Leave a comment

Pathway analysis of gene expression data – male reduced fertility / sterility

A group of animals that can breed and produce fertile offspring is one of the definitions of a species.

This means that the biological mechanisms of fertility and infertility are of interest not only to evolutionary biologists, but also to clinicians and of course to the wider public. At the Institute of Molecular Genetics in Prague, Prof. Jiri Forejt is studying what controls fertility in the hybrid offspring produced by the mating of mouse sub-species. He wanted to know why some male mice were infertile – he knew that genes in one particular genome region were important, but not how those genes influenced the expression of the rest of the genome.

This is where I was recruited into the team, to help with identifying the classes of genes disrupted in mice with reduced fertility. Scientists in his group had produced Affymetrix gene expression results from the testes of fertile, sub-fertile and infertile mice and I analysed these data genome-wide for differentially-expressed transcripts. Using the Broad Institute’s marvellous GSEA tool, I assessed the statistical evidence that specific Gene Ontology terms and pathways were over-represented and also whether differential genes were localised to particular genome regions. This analysis uncovered evidence that specific, functionally related sets of genes were over-represented in the expression data and helped to develop novel hypotheses about the causes of reduced fertility.

Posted in Pathway analysis, Target discovery | 1 Response

Target discovery in inherited muscle weakness

Muscle weakness can be caused by a rare inherited disease called myofibrillar myopathy. Gonzalo Blanco’s team found a mouse model of this disease and wanted to identify the underlying cause of the severe muscle weakness. Their aim was to discover potential therapeutic targets to translate into pre-clinical and clinical studies.

Before I became involved, the disease had been mapped to a large region of one chromosome and Dr Blanco’s team were planning to use conventional positional cloning methods to find the mutation. I proposed that a faster approach would be to use next-generation sequencing targeted at genes in the region. I designed a set of probes to enrich specific DNA fragments and I worked with a bioinformatician, Dr. Michelle Simon, to design a software pipeline to find and characterise mutations.

At the end of the design process, the pipeline was used to identify mutations in the muscle weakness mutants and predict that they altered the coding sequences of two genes; Myh4 and Pmp22. Two lines of evidence suggested that the mutation in Myh4, which codes for a muscle myosin protein, was the most likely cause of the weakness. Firstly, our colleagues found that mice carrying only the myosin mutation still had the trait and secondly, abnormal protein aggregates from affected mice contained large amounts of the myosin.

Scientists at the MRC’s Mammalian Genetics Unit have used the same approach, that Michelle Simon and I pioneered, to find mutations in other disease models.

Publication in Human Molecular Genetics

Testimonial from Dr. Gonzalo Blanco

Posted in Target discovery | Leave a comment
  • Connect with us

    Link to ourLinkedin
    Link to ourRss
    Link to ourTwitter