Abstract:
|
Finite mixture distributions are widely used to model and cluster heterogeneous data. Particular attention has been given to the use of mixture models with normal/t-component distributions and, more recently, with skew versions of such component distributions. In this talk, we focus on the case of component distributions that contain mixed-effects terms. They facilitate the modelling and clustering of feature vectors in situations there is some known a priori structure. In the case of replicated data, random-effects terms can be used to allow for correlation between repeated observations on the same entity and also to allow for correlations between entities in the same cluster. We give an example in the context of the clustering of gene profiles to improve the accuracy of the ranking and power of associated tests in the multiple testing of genes for no differential expression. Another example of a mixture model with mixed-effects components is given concerning the modelling and classification of multivariate observations in the case where there are multiple samples taken on objects with significant inter-object variation.
|