. 2022 Nov 21;13:7133. doi: 10.1038/s41467-022-34807-3

Table 2.

The optimization of DeepPROTACs model hyperparameters on validation set

Hyperparameters	Values	Average accuracy	AUROC
Batch size	1	77.15%^a	0.8246^a
	8	75.85%^a	0.8009^a
	16	75.85%^a	0.8183^a
	32	76.32%^a	0.8118^a
	64	76.21%^a	0.8212^a
	128	76.09%^a	0.8249^a
	256	75.62%^a	0.8246^a
Number of GCN layers	one	75.38%^a	0.7850^a
	two	77.15%^a	0.8246^a
	three	76.68%^a	0.8201^a
Pocket size	5 Å res around ligand	77.15%^a	0.8246^a
	10 Å res around ligand	76.33%^a	0.8114^a
	15 Å res around ligand	73.73%^a	0.8010^a
	All protein	73.97%^a	0.7780^a
Pooling layer	Max pooling	77.15%^a	0.8246^a
	Mean pooling	72.44%^a	0.7692^a
	Sum pooling	72.20%^a	0.7733^a
Inclusion of bond type encoding	No	77.15%^a	0.8246^a
Inclusion of bond type encoding	Yes	77.97%^a	0.8278^a
Optimized hyperparameters	use	77.46%^b	0.8531^b

Note: ^a metrics value on validation set, ^b metrics value on test set; bold means optimal parameters.