Prediction of Solvate Based on Graph Attention Network

In pharmaceutical development, the crystallization process is crucial for isolating and purifying Active Pharmaceutical Ingredients (APIs) using solvents. However, residual solvents left after drying can form solvates, altering crystal properties and potentially affecting drug quality. We developed a graph neural network model based on attention mechanisms to predict solvates. The results show that our model achieves SOTA results compared to the baseline model on most metrics for predicting solvates by capturing molecular interactions and key features. We have also set up a relevant website for the convenience of users: https://sanet.simm.ac.cn/. In order to better verify the performance of the model, we used SANet to predict the formation probability of NFD solvates in 24 solvents. We discovered a novel solvate and characterized its physicochemical properties.