The network architecture of GSScore. For a given protein–ligand conformation, GSScore divides it into multiple layers of shells with different distances, using the ligand molecule as the central reference point. Each shell includes protein atoms that are in proximity to the ligand, and a separate graph is constructed for each shell. These graphs are then individually inputted into a Graphormer to extract their respective feature vectors. The resulting feature vectors from all the graphs are concatenated into a single long vector, which is then passed through an MLP layer to predict the RMSD. Graphormer is an enhancement of the traditional Transformer model, incorporating spatial topological features and edge features based on the shortest path. This integration of graph topology and edge features helps to improve the predictive power of the model.