Which Method Is The Best Dissimilarity Measurement For Hierarchical Clustering Of Dna Methylation Data

Question

[Gangcai]

Dear all, I have performed hierarchical clustering for bisulfite DNA methylation data using two different dissimilarity methods: euclidean distance and pearson correlation by using R package pvclust. However,the tree structure based on those two methods are different. My question is which one should I use for hierarchical clustering of bimodal distributed DNA methylation data? Is there any published paper that have already compared different dissimilarity methods?

Thanks in advance.

[Michael]

As said often before (here and elsewhere) the attempt to recommend a single best method for data-mining is futile, given the lack of a gold standard to compare your results with. Clustering is exploratory and used for hypothesis generation, therefore the way to go is to apply many different methods (including other clustering methods: kmeans, Mclust) and try to evaluate the results in the light of your biological knowledge. Also, use e.g. GO analysis, pathway analysis and GSEA).

Answer 1

[Sean Davis]

I'd suggest converting these values to M-values, compute the distance metric, then display the beta values. Keep in mind that there is no "right" answer when clustering, so experimentation is necessary.