The document discusses the design of efficient Monte Carlo estimators for rare event probabilities in stochastic reaction networks using optimal importance sampling via stochastic optimal control. Two approaches are explored: a learning-based method for the value function and controls through stochastic optimization, and a Markovian projection-based method for solving reduced-dimensional Hamilton-Jacobi-Bellman equations. The goal is to address challenges like the curse of dimensionality while providing numerical experiments and results to support these methodologies.