Efficient computation by molecular competition networks

Abstract Most biomolecular systems exhibit computation abilities, which are often achieved through complex networks such as signal transduction networks. Particularly, molecular competition in these networks can introduce crosstalk and serve as a hidden layer for cellular information processing. Despite the increasing evidence of competition contributing to efficient cellular computation, how this occurs and the extent of computational capacity it confers remain elusive. In this study, we introduced a mathematical model for Molecular Competition Networks (MCNs) and employed a machine learning-based optimization method to explore their computational capacity. Our findings revealed that MCNs, when compared to their non-competitive counterparts, demonstrate superior performance in both discrete decision-making and analog computation tasks. Furthermore, we examined how real biological constraints influence the computational capacity of MCNs, and highlighted the nonnegligible role of weak interactions. The study suggested the potential of MCNs as efficient computational structures in both in vivo and in silico scenarios, providing new insights into the understanding and application of cellular information processing.