Choosing the best clustering method for a certain methylation data

Question

[Sophia]

Hello everyone,

I'm trying to choose the best clustering method for my methylation data (a set of DMRs) but I'm having a bit trouble to assess the power of biological interpretabilityof one clustering method to other. How can I tell that one clustering method gives more biologically meaningful and relevant results to the compared one for the same set of DMRs?

Any idea would help, many thanks in advance!

Answer 1

[yura.grabovska]

Something like ConsensusClusterPlus will allow you to assess cluster stability directly via the report it produces. Smiliar for something like NMF. More generally you can use silhouette as a measure of cluster stability but really there isn't always a great answer for which algorithm is better for X type of data when it comes to various 'omics.

When it comes to assigning biological relevance it's still up to you as the interpreter to decide whether the clustering has worked well, and given meaningful results. Ultimately the test comes in the form of subsequent validation. If you can extract some aspect of biology and validate it experimentally or in another dataset then that will strengthen your result. Fewer people will criticise you for using PAM instead of kmeans rather than making a biological interpretation that isn't well evidenced by the data.

Answer 2

[5heikki]

Affinity propagation is the answer to all my clustering needs