We used scPDB to divide the valid_data.set csv into a train-test set.
We increased the diversity and ease of training by dividing the scPDB by chain, but information about the binding site in multimers may be lost. The binding site binary is based on a 10Å radius from the center coordinates of the ligand.
PyTorch version: 1.10.0 PyTorch Geometric version: 2.0.2 ~