The field of molecular design and interaction prediction is rapidly advancing, with a focus on developing innovative methods for predicting molecular interactions, designing new molecules, and optimizing their properties. Recent developments have led to the creation of unified frameworks for activity cliff prediction, multi-type PTM site prediction, and test-time molecular optimization. These models have shown significant improvements over existing approaches, achieving state-of-the-art performance on various benchmarks. Notably, the integration of multi-modal information, such as protein sequences, structures, and evolutionary information, has been a key factor in the success of these models. Additionally, the use of large language models and generative models has enabled the design of proteins with desired functions and the prediction of antibody-antigen binding affinity. The development of new datasets and evaluation frameworks has also facilitated the advancement of this field. Some noteworthy papers in this area include MTPNet, which achieved an average RMSE improvement of 18.95% on activity cliff prediction datasets, and Llama-Affinity, which showed significant improvement over existing methods for antibody-antigen binding affinity prediction. Overall, these advances have the potential to accelerate compound optimization and design, and to improve our understanding of molecular interactions and their role in various biological processes.