Table 5. Comparison of Different Molecular Representationsa.
assay code | MOE | MACCS | MOE + fold. ECFP4 | nonfolded ECFP4 | folded ECFP4 |
---|---|---|---|---|---|
A | 0.91 | 0.90 | 0.93 | 0.93 | 0.93 |
B | 0.65 | 0.64 | 0.68 | 0.70 | 0.70 |
C | 0.66 | 0.67 | 0.68 | 0.69 | 0.69 |
D | 0.59 | 0.60 | 0.63 | 0.65 | 0.62 |
E | 0.86 | 0.84 | 0.90 | 0.93 | 0.94 |
F | 0.86 | 0.84 | 0.87 | 0.87 | 0.87 |
G | 0.58 | 0.56 | 0.60 | 0.57 | 0.58 |
H | 0.76 | 0.73 | 0.76 | 0.77 | 0.77 |
I | 0.85 | 0.86 | 0.87 | 0.90 | 0.89 |
J | 0.90 | 0.92 | 0.93 | 0.93 | 0.93 |
Reported are AUC values for prediction of 10 assays (codes A–J) in matrix 1 using per-target RF models on the basis of different molecular representations, including 192 two-dimensional (2D) descriptors from the Molecular Operating Environment (MOE), 166 MACCS structural keys, the folded and unfolded version of ECFP4, and the combination of MOE descriptors and folded ECFP4 (MOE + fold. ECFP4). For each assay, best results are indicated in bold.