Finite Mixture Models in Biostatistics
MetadataShow full item record
Mixture models are powerful tools for density estimation and cluster and discriminant analyses. They have enjoyed widespread popularity in biostatistics, biomedicine, medical imaging, and genetics, among many other applied fields. The mixture model framework provides a formal but convenient and flexible approach to model complex heterogeneous datasets such as those that typically arise in biological studies. This chapter discusses two interesting applications of mixture models in biostatistics, namely, the analysis of cytometry data and of microarray data. We begin with a brief overview of mixture models and a general discussion of recent advances in this area, focusing on trends that are relevant to biostatistics and health science. In particular, we consider techniques that address challenges in analyzing large biomedical datasets, including dimension reduction, the handling of asymmetric and nonnormal clusters, and accounting for inter- and intracluster variations. These are demonstrated via the EMMIX-JCM and EMMIX-contrasts procedures, which are based on random-effects skew mixture models and linear mixed-effects mixture models, respectively. In several applications of EMMIX-JCM to flow cytometric datasets, we illustrate how mixture models can automate the segmentation of cells in samples, align clusters across samples, build batch templates, and predict the labels for new samples. Further illustrations are given using EMMIX-contrasts on real and simulated microarray datasets to showcase the effectiveness of mixture models in clustering gene expression data, ranking genes, and controlling false discovery rate.
Disease Modelling and Public Health, Part A
Medical and Health Sciences not elsewhere classified