Abstract
Biomolecular networks underlie both natural biological processes and engineered cellular technologies, from intracellular regulation and ecological dynamics to biomanufacturing, smart therapeutics, and cell-based diagnostics. However, designing chemical reaction networks (CRNs) that implement a desired dynamical function remains a challenging task. Although candidate networks can be evaluated by simulation, the inverse problem of discovering networks from behavioral specifications remains difficult. It requires navigating vast spaces of topologies and kinetic parameters governed by nonlinear and potentially stochastic dynamics. Here, we introduce GenAI-Net, a generative artificial intelligence framework that automates CRN design by coupling reaction proposal to simulation-based evaluation defined by a user-specified objective. GenAI-Net efficiently produces topologically diverse solutions across design tasks, including dose-response shaping, complex logic gates, classifiers, oscillators, habituation, robust perfect adaptation, and noise reduction in stochastic settings. By turning specifications into families of circuit candidates, GenAI-Net provides a route to programmable biomolecular circuit design and accelerates translation from desired function to implementable mechanisms.