Most similarity techniques are optimized for finding nearest neighbors or for generating clusters. But what if you need to pick most dissimilar compounds or to rank a compound list by dissimilarity?

I believe that component "Cluster Molecules" does the job if the CenterSelectionMethod parameter is set to "Maximum Dissimilarity" and parameter NumberOfClusters set to the number of molecules. But the result lists differ when changing the NumberOfClusters, and there is no option to check for the calculated dissimilarity e.g. to sort the molecules.

So, what is the best dissimilarity method which can be applied from shipped components?