Table 1 Performance comparison of different training settings by the average and standard deviation on the internal test set of 40 protein targets
From: Deep scaffold hopping with multimodal transformer neural networks
Metrics | Models | |||
|---|---|---|---|---|
Single-task | DeepHop-noGNN | DeepHop-noProtein | DeepHop | |
Success rate (%) | 27.5(15.9) | 61.8(18.6) | 58.9(20.9) | 65.2(17.5) |
Constraint success (%) | 15.5(14.7) | 40.6(20.9) | 34.6(19.7) | 43.7(21.0) |
Improvement | 0.53(0.31) | 0.92(0.27) | 0.64(0.28) | 0.97(0.24) |
Validity (%) | 12.9(6.3) | 94.4(2.7) | 92.7(3.8) | 95.7(3.8) |
Uniqueness (%) | 8.7(5.5) | 74.6(11.1) | 88.2(8.8) | 76.4(9.3) |
Novelty (%) | 99.0(0.9) | 99.5(0.5) | 99.6(0.3) | 99.4(0.5) |