Deep Reinforcement Learning for Scalable Control of Boolean Models in the Context of Cellular Reprogramming

This study introduces pbn-STAC, a novel computational framework based on deep reinforcement learning, designed to identify reprogramming strategies for large Boolean and Probabilistic Boolean Network models in cellular reprogramming. The framework addresses the challenge of computational complexity by introducing pseudo-attractors, which represent stable states corresponding to cell types or phenotypic states. pbn-STAC effectively finds control targets and strategies to transition networks from source to target states by intervening in intermediate attractor states observable in wet-lab practice. The framework's performance is evaluated on various models, including a biological case study of immune response against B. bronchiseptica infection, demonstrating its effectiveness compared to exact, optimal solutions. This work contributes to developing scalable control methods for large gene regulatory network models.