Dear Michael,

thanks for the great answer. You also rise the issue whether one should consider only biologically relevant regions, as known from literature, from irrelevant ones. This is not unbiased though. My question is maybe even easier in reality: if you have several subclones and you want to cluster them according to the greatest possible similarity, in order to infer which is deriving from which, wouldn't it just be easier generate a matrix of amp/del status along chromosomes and then cluster everything?

Have you tried on the field of symbolic data analysis? They have defined several dissimilarity indices for many different cases. Authors like Diday, De Carvalho, Verde, Lechevallier could have studied the problem of dissimilarity definition in your context. After that, the clustering method will depend on what do you want: a tree structure or a partition structure. No method is perfect, but a good one should verify a Huygens-like theorem to satisfy therepresentability problem by means of a "centroid".

For one cell, yes the copy number is discrete value, but when you consider a "sample", e.g. a tumor block or a cultured cell line, there is always heterogeneity between cells, so to me the copy number is always a continues value, the average of numerous cells. On the other hand, the copy number may not follow Gaussian distribution, which may lead to different consideration when you calculate distance matrix.