Bayesian classifiers with consensus gene selection: a case study in the systemic lupus erythematosus

Abstract

Within the wide field of classification on the Machine Learning discipline, Bayesian classifiers are very well established paradigms. They allow the user to work with probabilistic processes, as well as, with graphical representations of the relationships among the variables of a problem. Bayesian classifiers assign the corresponding predicted class of a certain pattern as the one that has the highest a posteriori probability. This a posteriori probability is computed by means of the Bayes theorem in conjunction with assumptions about the density of the patterns conditioned to the class. In this work three of these classification paradigms are applied to a DNA microarray database of control, systemic lupus erythematosus and antiphospholipid syndrome samples. The number of genes from which the models are induced is considerably reduced by means of a novel consensus filter gene selection technique. Combining a nonparametric bootstrap resampling technique and the k dependence Bayesian classifier paradigm, we propose a new method to obtain gene interaction networks of high reliability. These gene networks can be seen as a tool to study the relationships among the genes of the domain. In fact, some of the previous knowledge about both pathologies is confirmed by the new approach.