Pareto-Optimal Interventions in Gene Regulatory Networks using Signal Temporal Logic

This paper presents a framework for identifying Pareto-optimal intervention policies in Gene Regulatory Networks (GRNs), addressing both the complexity and uncertainty inherent in biological systems. Existing methods, such as dynamic programming and reinforcement learning, focus primarily on average intervention performance. However, practical interventions in systems biology must account for multiple competing objectives, including worst-case performance, response time, intervention frequency, and long-term system stability. To model the stochastic dynamics of GRNs, this paper employs Boolean networks with perturbations (BNp) and formulates the intervention problem as a constrained multi-objective optimization task. Signal Temporal Logic (STL) is leveraged to evaluate policies, particularly focusing on maximizing intervals free of therapeutic side effects and minimizing repeated interventions. Our method generates a Pareto-optimal set of policies, providing biologists with a flexible range of solutions tailored to specific experimental needs. Numerical experiments demonstrate the effectiveness of our approach in achieving robust and efficient intervention performance.