A machine learning approach to infer gene interaction networks

Gene interaction networks play an important role in the area of microarray analysis. The reconstruction of those networks from gene expression profiles is a relevant task and several statistical approaches have been used, during the past, to address this problem. Among them the correlation based methods are probably the most employed. They define interactions among genes that have a similar trend across different samples. We propose a Machine Learning based approach to create gene interaction networks in order to include relationships that are hardly recognizable by those traditional statistic algorithms. First we define a general pipeline to infer this interaction networks, then we compare it with the well known co-expression approach. We show that our networks are different from the co-expression ones in terms of topology properties and biological knowledge, thus they represent a new way to discover gene-gene interactions.