The paper introduces the Graph Convolutional Policy Network (GCPN), a novel approach for goal-directed molecular graph generation. GCPN is designed to optimize molecular properties while adhering to underlying chemical rules, such as valency constraints. The model leverages graph representation learning, reinforcement learning, and adversarial training to achieve this. GCPN is trained to optimize domain-specific rewards and adversarial loss through policy gradient, operating within an environment that incorporates these rules. Experimental results show that GCPN outperforms state-of-the-art baselines by achieving 61% improvement in chemical property optimization and 184% improvement in constrained property optimization tasks. The method is evaluated on three tasks: molecule property optimization, property targeting, and conditional property optimization, demonstrating its effectiveness in generating molecules with desired properties while maintaining high validity and resemblance to realistic molecules.The paper introduces the Graph Convolutional Policy Network (GCPN), a novel approach for goal-directed molecular graph generation. GCPN is designed to optimize molecular properties while adhering to underlying chemical rules, such as valency constraints. The model leverages graph representation learning, reinforcement learning, and adversarial training to achieve this. GCPN is trained to optimize domain-specific rewards and adversarial loss through policy gradient, operating within an environment that incorporates these rules. Experimental results show that GCPN outperforms state-of-the-art baselines by achieving 61% improvement in chemical property optimization and 184% improvement in constrained property optimization tasks. The method is evaluated on three tasks: molecule property optimization, property targeting, and conditional property optimization, demonstrating its effectiveness in generating molecules with desired properties while maintaining high validity and resemblance to realistic molecules.