Table 2. Pose Classification Compared to Other Methods Across Different Train/Test Splitsa.
| ROC
AUC (PR AUC) |
|||
|---|---|---|---|
| method | Uniprot | seqsim70 | seqsim50 |
| ANPR | 0.93 (0.47) | 0.90 (0.45) | 0.89 (0.39) |
| smina docking | 0.82 | 0.82 | 0.81 |
| Cornell et al.58 | 0.86 | ||
| Ragoza et al.60 | 0.815c | ||
| Lim et al.28 | 0.94b | ||
a
Baseline random performance is 0.5 for ROC AUC and 0.05 for PR AUC, reflecting the presence of 5% true positive good poses in the test set.
b
In contrast to other methods in this table, the model of Lim et al. was trained simultaneously on poses and activity.
c
Three-fold cross-validation on 90% sequence identity.