Analyzing Boolean Networks Through Unsupervised Learning

Abstract

AbstractBoolean networks are typically used as simple models of gene regulatory networks. We use a particular class of Boolean networks called threshold Boolean networks defined by a weight matrix, a threshold vector, and an updating mode in this work. We consider the reconstruction of synthetic threshold Boolean networks that contain the same fixed points as the Mendoza and Alvarez-Buylla network of flower development by using an evolution strategy. We propose a characterization by computing topological and dynamical features of the inferred synthetic networks and then applying machine learning, particularly unsupervised learning techniques, to analyze these networks. We discover how these networks are clustered and what features are relevant to discriminate the cluster containing the Mendoza and Alvarez-Buylla network from all the other clusters.