Table 2 Models’ performance averaged over the 46 targets for the given algorithm and validation approach
From: ProfhEX: AI-based platform for small molecules liability profiling
Validation | Algorithm | R | R2 | RMSE |
|---|---|---|---|---|
External | MLR | 0.62 (0.1) | 0.35 (0.2) | 1.08 (0.17) |
GB | 0.84 (0.05) | 0.68 (0.1) | 0.69 (0.08) | |
RF | 0.81 (0.07) | 0.64 (0.11) | 0.79 (0.1) | |
Bootstrap | MLR | 0.66 (0.1) | 0.43 (0.13) | 1.02 (0.14) |
GB | 0.79 (0.07) | 0.63 (0.11) | 0.77 (0.11) | |
RF | 0.79 (0.07) | 0.6 (0.11) | 0.85 (0.1) | |
fivefold cross validation | MLR | 0.65 (0.11) | 0.42 (0.15) | 1.02 (0.14) |
GB | 0.83 (0.06) | 0.67 (0.1) | 0.71 (0.09) | |
RF | 0.82 (0.06) | 0.65 (0.1) | 0.79 (0.1) | |
Y-scrambling | MLR | 0.0 (0.1) | 0.46 (0.14) | 0.99 (0.13) |
GB | 0.0 (0.02) | 0.22 (0.09) | 1.49 (0.28) | |
RF | 0.0 (0.02) | 0.05 (0.01) | 1.37 (0.26) |