Table 2.

Overview of the number of protein sequences in the training, validation and test set of the seven downstream protein structural prediction tasks.

Dataset	SS3	SS8	BUR	ASA	PPI	EPI	HPR
Training set	8 803	8 803	8 803	8 803	324	179	2 949
Validation set	1 102	1 102	1 102	1 102	81	45	738
Test set	1 102	1 102	1 102	1 102	137	56	1 230

The prediction tasks secondary structure in three (SS3) and eight (SS8) classes, buried residues (BUR) and absolute solvent accessibility (ASA) are based on the same dataset. Furthermore, we included the prediction tasks protein-protein interaction interfaces (PPI), epitope interfaces (EPI) and hydrophobic patches (HPR).