Table 1:

RBFE performance after ablating different components of the network on the 1 additional ligand set to determine their utility in the full network. Parentheses indicate the ±1 standard deviation of the 25 train/test versions. Bold indicates that it is not significantly different from the Standard model (p > 0.005).

Ablation	Pearson’s R	RMSE (pK)	MAE (pK)
Standard	0.553(±0.0233)	1.11(±0.0309)	0.82(±0.0187)
No LΔΔG	0.551(±0.0202)	1.12(±0.0248)	0.829(±0.0179)
No LΔG	0.459(±0.0238)	1.27(±0.0289)	0.945(±0.0182)
No LRotation	0.556(±0.0188)	1.11(±0.0233)	0.819(±0.0162)
No LConsistency	0.536(±0.021)	1.14(±0.0356)	0.842(±0.0186)
No LΔΔG, LConsistency	−0.0576(±0.136)	1.24(±0.0143)	0.908(±0.0144)
No LΔG, LConsistency	0.456(±0.0231)	1.28(±0.0319)	0.95(±0.0233)
Concatenation	0.554(±0.0134)	1.11(±0.0223)	0.821(±0.0174)
No Siamese Network	0.5(±0.0347)	1.15(±0.0362)	0.854(±0.021)
Subtraction, Single-order	0.512(±0.0213)	1.17(±0.0213)	0.877(±0.0151)
Concatentation, Single-order	0.476(±0.023)	1.21(±0.0253)	0.907(±0.0182)