Method
|
AUC (SD
−
)
|
AUC (SD
+
)
|
AUC (OB)
|
|---|
NC | 0.993 | 0.844 | 0.919 |
C
O
S
O1
| 0.979 | 0.987 | 0.994 |
C
O
S
O2
| 0.971 | 0.978 | 0.992 |
C
O
S
O1 w | 0.98 | 0.987 | 0.996 |
C
O
S
O2 w | 0.971 | 0.973 | 0.996 |
M
W
M
O1
| 0.98 | 0.982 | 0.996 |
M
W
M
O2
| 0.972 | 0.974 | 0.996 |
M
W
M
O1 w | 0.98 | 0.981 | 0.996 |
M
W
M
O2 w | 0.972 | 0.969 | 0.997 |
- The AUC scores are computed from the TPR and FPR of the different measures. The scores reflect the quality of the COS, MWM and NC measures for protein family classification. An AUC score of 1 corresponds to a perfect classification of the dataset. All methods produce a very good AUC score, a small general advantage can be observed for the methods using an order 1 parameter. Cosine methods have better performances on the SD + dataset and the MWM methods perform generally better on the SD − dataset and on the OB dataset. Using the weighted version of the COS or MWM measure only improves the performance on the OB dataset.
