Clustering data with multiple cluster structures: a model-based perspective