SAVJTNNGAN: More Specific Information for Graph Structure Translation

Graph structure translation is an important re-search problem applicable in various scenarios, where graph translation models serve as advantageous tools for relationship prediction and network optimization. The application of graph translation in the field of drug molecules is a prominent area of deep learning implementation, focusing on optimizing target properties while maintaining a certain similarity between evolved molecules. To enhance the success rate of translation, we propose a new molecular translation model based on the VJTNNGAN model, which integrates SAGAN to optimize feature extraction using a message passing network combined with attention mechanisms. Additionally, we enhance the performance of the graph decoder by incorporating graph decoding information into GAN network training. Finally, we validate the translation performance of the SAVJTNNGAN network through testing on relevant datasets. And the translation success rate has improved on both the QED and Drd2 datasets.