Dear Renzo,

The function find.clusters displays a graph of cumulated variance explained by the eigenvalues of the PCA. Apart from computational time, there is no reason for keeping a small number of components; so you can keep all the information, specifying it to keep 120 PCs in your case. The function then displays a graph of BIC values for increasing values of k. Your graph shows a clear decrease of BIC until k = 3 clusters, after which BIC increases. This indicates that 3 clusters should be retained; the ’best’ BIC is often indicated by an elbow in the curve of BIC values as a function of k. In practice, the choice is often trickier to make for empirical datasets; remember that clusters are merely tools used to summarise and understand the data. There is no ”true k”, but some values of k are better, more efficient summaries of the data than others. Genetic clusters can be biologically meaningful structures and reflect interesting biological processes, but they are still models (Jombart & Collins, 2015).  

Thank you so much Angelica for your answer.

Dear Renzo and Angelica

I have the same problem with find cluster. For my case i kept all PCs which are 140. But i can not figure out the true K. Please help

He everyone,

I also have doubts about the correct number of clusters to choose. I have a dataset with a very soft population structure differentiation, although some structuring exists for at least two different populations. This is the graph I get from find.clusters after retaining 450 PC's. Does this means that find.clusters is not finding any structure?

Thanks in advance for your help.

Hi Natalia!

See you also here.

Did you tried to retain all the PCs? What's the max number of K you ran it? You can also try to use the optim.a.score() to have an idea of the suggested number of PCs to retain for your analysis. Yes, it seems that the function is not finding anything. What about the outliers?

Does anyone do DAPC for morphomertric mesurements?