A sequential clustering algorithm with applications to gene expression data.

摘要

Microarrays are part of a new class of biotechnologies which allow the monitoring of expression levels for thousands of genes simultaneously. Gene profile data come from experiments that investigate the behavior of genes over several time points. Biologists are interested in these gene expression profiles because it is believed that genes in the same functional pathway have similar profiles of gene expression.; In the analysis of data from microarray experiments, most of the unsupervised learning processes involve three steps: standardization, defining a dissimilarity measure, and applying a clustering algorithm. We will discuss the issues involved in these steps, and we will propose new methods. We will discuss the problems of current clustering algorithms and propose a new algorithm, the sequential clustering algorithm. This algorithm finds clusters sequentially based on a gaussian model. The algorithm does not require the specification of the number of clusters and allows for sporadic objects.; We will discuss a semiparametric mixture model which is motivated by the sequential clustering algorithm. Two estimators for the mixing proportion in semiparametric mixture model are proposed, and their properties are investigated using simulations.; A new dissimilarity measure that takes into account the time order and the time distance between experiments will be introduced. We will discuss the performance of various distances in clustering using the Asymptotic Discriminating Measure (ADM) and show that the new dissimilarity measure has always higher ADM than the Euclidean distance. The comparison of distances in small samples will be also discussed. We will introduce a sequential clustering algorithm with the new dissimilarity measure and investigate its performance.