An application of the non-rejection rate in meta-analysis

Learning of transcriptional regulatory interactions using microarray data is an important and challenging problem in bioinformatics. Several solutions to this problem have been proposed both in a model based statistical approach and in an unsupervised machine learning approach. In a model based approach a very popular choice are Gaussian graphical models where it is assumed that microarray data form an i.i.d. multivariate normal sample. In this framework, Castelo and Roverato (2006) introduced a quantity that they called the non-rejection rate that can be used to learn Gaussian graphical models when the sample size is smaller than the number of variables. Here we present an application of the non-rejection rate, in an unsupervised learning approach, to a compendium of data from different microarray experiments and shown that it provides competitive performance with respect to other widely used methods.