Table 1. Performance Measures in Predicting Pgp Substrates by Using the Data Set of Li et al.16.
| method | parameters | accuracy | sensitivity | specificity | κ | coverage |
|---|---|---|---|---|---|---|
| Traininga | ||||||
| v-NN | HC (h = 0.3; d0 = 1.0)b | 0.71 | 0.78 | 0.65 | 0.42 | 1.00 |
| v-NN | HA (h = 0.6; d0 = 0.6) | 0.77 | 0.78 | 0.75 | 0.53 | 0.60 |
| BC | ECFP10 + 8MPc | 0.72 | 0.65 | 0.79 | 0.44 | 1.00 |
| Test | ||||||
| v-NN | HC (h = 0.3; d0 = 1.0) | 0.76 | 0.80 | 0.71 | 0.51 | 1.00 |
| v-NN | HA (h = 0.6; d0 = 0.6) | 0.81 | 0.82 | 0.78 | 0.60 | 0.70 |
| BC | ECFP10 + 8MP | 0.73 | 0.66 | 0.81 | 0.47 | 1.00 |
| Randomized Training Datad | ||||||
| v-NN | HC (h = 0.3; d0 = 1.0) | 0.50 | 0.50 | 0.50 | 0.01 | 1.00 |
| v-NN | HA (h = 0.6; d0 = 0.6) | 0.52 | 0.50 | 0.54 | 0.04 | 0.65 |
a
Performance in a 10-fold cross validation.
b
v-NN parameters (smoothing factor, h, and Tanimoto-distance threshold, d0).
c
Bayesian classifier that employs ECFP10 fingerprints and eight molecular properties (MPs).
d
Training set compounds for Pgp were randomly assigned as substrates or nonsubstrates.