The field of molecular drug discovery is moving towards the development of more efficient and scalable methods for generating novel molecules with desired properties. Recent research has focused on combining generative models with active learning strategies and prior knowledge to improve the quality and diversity of generated molecules. This has led to the development of new frameworks and models that can generate molecules with high binding affinity and desired pharmacological properties. Notable papers in this area include VECTOR+, which introduces a valid-property-enhanced contrastive learning framework for targeted optimization and resampling, and PAFlow, which proposes a novel target-aware molecular generation model featuring prior interaction guidance and a learnable atom number predictor. Additionally, BALD-GFlowNet and PoolPy have been proposed as scalable and flexible methods for active learning and group testing design, respectively.