Learning binary threshold networks for gene regulatory network modeling

Inspired by the resent trend of binary neural net-works, where weights and activation thresholds are represented using 1 and-1 such that they can be stored in 1-bit instead of full precision, we explore this approach for gene regulatory network modeling. An evolutionary computation approach to learn binary threshold networks is presented. In particular, we consider differential evolution and particle swarm optimization. We test our method by inferring binary threshold networks of a regulatory network of Quorum sensing systems in bacterium Paraburkholderia phytofirmans PsJN. We present results for weights having only 1 and-1 values, and consider different activation thresholds. Full binary threshold networks were found with minimum error (2 bits), whereas when the binary restriction is relaxed for the activation thresholds, networks with 0 bit error were found.