I'm trying to find a suitable method for detection of outliers from a PCoA output. This analysis is used to visualize the results from a distance matrix between a set of sample applying the Chemical Structure and compositional similarity (https://github.com/madeleineernst/q2-cscs#2-compute-the-chemical-structural-and-compositional-dissimilarity-for-a-real-world-dataset>).

My interest are those sample which are outliers, or away from the clusters.

Can someone give me some advices if is possible to obtain the ourliers from a distance matrix, or form the PCoA cosidering all the coordinates, or as many as possible?


  • $\begingroup$ PcoA is a simplest form of classic MDS. It turms your nxn distance matrix into a nxm dataset. Now, if m is large (approaches n) then both the matrix and the dataset are just two "equivalent" forms of the data. So it makes no difference where you will be detecting outliers - here or there. But if m is small, then the dataset is the dim.-reduced form of the original data, and there appeares a question whether this reduction has kept or has lost the info about outliers. And it is hard to answer the question. But you may still search for outliers right in the initial dist. matrix. $\endgroup$ – ttnphns Sep 23 '20 at 15:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.