I would like to find out how to combine gene set enrichment analysis with hierarchical clustering. The motivation for this combination is that potentially too many gene-set symbols for leukemia may pass the p-value significance threshold. We are using an R-package for fast preranked gene set enrichment analysis (GSEA) with the URL, https://github.com/ctlab/fgsea.

Subsequent to the fast gene set enrichment analysis on a ranked list of gene symbols with different expression t-statistics, we specifically need to identify gene-set symbols belonging to a common functional group using the R language hclust function intended for supervised hierarchical clustering with a Euclidean distance between features in a timely manner, for example 15 or fewer minutes.

Quoting from Alan Moses 2017 book, "Statistical Modeling and Machine Learning for Molecular Biology" , "Clustering is meant for exploratory data analysis and therefore doesn't really have a strong framework for hypothesis testing".A new R package ClusterProfiler provides enrichment analysis of gene clusters as reported in this URL, http://guangchuangyu.github.io/2015/05/use-clusterprofiler-as-an-universal-enrichment-analysis-tool/

The fact that most quality measures found in the literature have been conceived to evaluate non-overlapping clusterings, even when most real-life problems are better modeled using overlapping clustering algorithms is analyzed in detail in the following paper and University of Texas Ph.D thesis

Academic paper: CICE-BCubed: A New Evaluation Measure for Overlapping Clustering Algorithms. Available from: https://www.researchgate.net/publication/260421976_CICE-BCubed_A_New_Evaluation_Measure_for_Overlapping_Clustering_Algorithms [accessed Apr 16, 2017].



As I am a novice to this type of bioinformatics research, please correct any inaccuracies in my problem statement.

  • $\begingroup$ The article, What to Do When K-Means Clustering Fails: A Simple yet Principled Alternative Algorithm with the URL, journals.plos.org/plosone/article?id=10.1371/… discusses MAP-DP which is a more flexible clustering technique using a probabilistic model. $\endgroup$ – Frank Apr 17 '17 at 8:51

You can try Enrichr for enrichment analysis. On 'Results' page you can use Clustergram view. Or you can upload your results to Clustergrammer itself.

  • $\begingroup$ Thank you for your nice answer. Is it possible to measure and refine the quality of hierarchical clustering results generated by the R language function hclust() using Euclidean distance by automatically computing p-values for all clusters contained in the clustering of original data. Could we use the paper "Multiscale mutation clustering algorithm identifies pan-cancer mutational clusters associated with pathway-level changes in gene expression", journals.plos.org/ploscompbiol/article?id=10.1371/… to refine clustering analysis? Thank you. $\endgroup$ – Frank Apr 16 '17 at 19:09
  • $\begingroup$ Is it mathematically correct to use the R function for combining gene set enrichment analysis p-values from github.com/ctlab/fgsea. in this URL, biostars.org/p/211307, to calculate the quality of a hclust() clustering result tree? Thank you. I am awarding reputation points to your nice answer. $\endgroup$ – Frank Apr 16 '17 at 19:15
  • $\begingroup$ How does your answer address overlapping clustering? Thank you. $\endgroup$ – Frank Apr 17 '17 at 3:48

A rising young M.I.T professor told me just now that the volcano plot with 1og(p value) vertical axis can in many cases differentially visualize overlapping clusters where p values come from gene set enrichment analysis,


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.