Learning topology-preserving embedding for gene interaction networks

Learning topology-preserving embedding for gene interaction networks

Abstract

Understanding functional aspects of genes or proteins is crucial to providing insights into underlying mechanisms for different health and disease conditions. Gene interaction networks model these complex biological phenomena and provide rich information to infer functional relationships among genes. However, comprehensive representation of topological properties of gene interaction networks, to achieve more accurate inference of new gene interactions, still remains a challenge. Here, we describe a deep neural network architecture to learn lower dimensional representation for each gene, by preserving direct and indirect topological proximity between genes, that characterizes the topological context of each gene. These representations can be plugged into off-the-shelf machine learning methods to derive deeper insights into the structure of gene interaction networks and also functional insights about genes. We compare our method to the state-of-the-art machine learning methods and demonstrate significant improvement in predicting gene interactions.

Publication
17th European Conference on Computational Biology (ECCB) (Poster)