I have N samples and each of them is characterized by a vector of M descriptors. I can compute the pairwise distances between samples by means of some distance measure (e.g. cosine). Now let's say that I compute the descriptors again, under different conditions. I want to check if the differences between samples remain (more or less) constant. For example, are there samples that were similar and now, under the new condition, are distant? One way could be computing the correlation between the dissimilarity matrices. Are there other statistical tools to determine if different conditions impact on the relationships between samples?