Efficient Inverse Design of Stochastic Molecular Circuits

Existing methods for the inverse design of biological regulatory circuits - such as gene networks - often disregard stochastic effects, relying on problem-specific techniques. We present a novel, general gradient-based optimization approach that addresses this limitation. Our framework enables parameter fitting of exact Stochastic Simulation Algorithms (SSAs), including the Gillespie algorithm, without resorting to smooth approximations. This method preserves the integrity of rare events and offers superior scalability compared to recent alternatives. To make these tools widely accessible, we developed a flexible JAX-based library for stochastic simulations and optimization, supporting various SSA algorithms. We demonstrate the effectiveness of our approach through optimization examples across diverse stochastic systems, including those with information processing in gene expression and spiking neural networks. This work has significant implications for synthetic biology, developmental biology, and tissue engineering, offering a powerful tool for designing and controlling complex biological systems under stochastic conditions. Our framework paves the way for more accurate and efficient inverse design of biological circuits, potentially accelerating progress in fields ranging from biomedicine to biocomputing.