Tight clustering for large datasets with an application to gene expression data

This article proposes a practical and scalable version of the tight clustering algorithm. The tight clustering algorithm provides tight and stable relevant clusters as output while leaving a set of points as noise or scattered points, that would not go into any cluster. However, the computational limitation to achieve this precise target of tight clusters prohibits it from being used for large microarray gene expression data or any other large data set, which are common nowadays. We propose a pragmatic and scalable version of the tight clustering method that is applicable to data sets of very large size and deduce the properties of the proposed algorithm. We validate our algorithm with extensive simulation study and multiple real data analyses including analysis of real data on gene expression.