Topological Approaches to Understanding Genetic Networks.

Abstract

In this study, we present methods of network analysis from both statistical and topological points of view to enhance our understanding of the robustness of the networks and the gene – gene interactions they exhibit. Using different strategies, networks were learned from a single cell dataset in which expression values were collected during various stages of cell development. The bnlearn algorithm was used to construct directed and acyclic Bayesian networks for statistical analysis which revealed greater homogeneity of activating interactions than that of inhibiting interactions. Because topological data analysis relies of cycles formed by edges in the network which are – by construction – not present in the Bayesian networks, new networks were learned by computing entropy, mutual information, and distances manually. Performance of persistence homology with the TDA algorithm revealed topological features which rendered the networks topologically distinct.