Clustering High Dimensional Discrete Sequences Using Mixture Trees

Abstract

An ancestral mixture model is proposed for clustering discrete multivariate sequences. This model has a natural relationship to the coalescent process of population genetics. The sieve parameter in the model plays the role of time in the evolutionary tree of the sequences. By sliding the sieve parameter, one can create a hierarchical tree that estimates the population structure at each fixed backward point in time. In this seminar, some theoretical and computational properties of the ancestral mixture model will be presented, which will include the assessment of goodness-of fit. Promising results from a case study will also be shown. Furthermore, potential applications will be discussed.