Programmable Biochemical Logic Computation Through Repurposing Natural bZIP Protein Interaction Networks

Protein dimerization networks can implement biochemical logic, but their practical realization has largely relied on designing or optimizing protein-protein interaction affinities, creating a major barrier to experimental implementation. Here, we ask whether computation can instead be extracted from interaction networks already generated by evolution. We repurpose experimentally characterized basic leucine zipper (bZIP) interactomes from Nematostella vectensis, Ciona intestinalis, and Homo sapiens as programmable biochemical logic substrates while keeping all measured binding affinities fixed. Using mass-action equilibrium modeling coupled with genetic optimization, we identify accessory monomer concentrations and output dimers that reconfigure these natural networks to perform Boolean operations. Unengineered bZIP clusters computed up to 15 of the 16 possible two-input Boolean functions, including nonlinearly separable XOR and XNOR operations, and up to 13 functions within a 17-gate three-input benchmark. Computational versatility depended on network architecture and input selection, revealing design constraints inherent to natural interactomes. Stochastic perturbation analysis further showed that many optimized configurations preserve logic-state separation under substantial concentration variability. Finally, a nondimensional scaling framework provides a route for translating simulated concentration regimes toward cell-free implementation. Together, these results demonstrate computationally that natural protein interaction networks contain latent, evolutionarily encoded computational capacity that can be accessed through concentration tuning rather than de novo protein engineering.