Table 4A.
Summary of binding ligand prediction by the 2D pocket model using Pseudo-Zernike moments on Kahraman dataset.
| Descriptora) | Rank | AMP | ATP | FAD | FMN | GLC | HEM | NAD | PO4 | STR | Averageb) | Totalc) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| pocket shape + pocket size (G) | Top1 | 11.1(%) | 57.1 | 60.0 | 0.0 | 60.0 | 50.0 | 33.3 | 100.0 | 0.0 | 41.2 | 51 |
| Top3 | 77.8(%) | 100.0 | 80.0 | 33.3 | 100.0 | 68.8 | 86.7 | 100.0 | 40.0 | 76.3 | 82 | |
| Average distance | 5.01 | 3.8 | 5.02 | 7.28 | 4.4 | 4.47 | 5.73 | 30.6 | 3.78 | 4.73 | - | |
| Elec. Potential + pocket size (E) | Top 1 | 11.1 | 0.0 | 70.0 | 0.0 | 20.0 | 0.0 | 46.7 | 100.0 | 0.0 | 27.5 | 36 |
| Top 3 | 33.3 | 64.3 | 80.0 | 0.0 | 40.0 | 56.2 | 66.7 | 100.0 | 20.0 | 51.2 | 62 | |
| Avg. distance | 11.64 | 11.14 | 10.14 | 16.27 | 17.35 | 11.06 | 9.14 | 7.63 | 15.16 | 12.17 | - | |
| Pocket shape+ size + electrostatic potential (G+E) | Top 1 | 22.2 | 57.1 | 70.0 | 0.0 | 40.0 | 50.0 | 33.3 | 100.0 | 0.0 | 41.4 | 52 |
| Top 3 | 77.8 | 92.2 | 80.0 | 50.0 | 100.0 | 86.2 | 86.7 | 100.0 | 40.0 | 75.9 | 80 | |
| Avg. distance | 5.88 | 4.76 | 5.52 | 8.3 | 5.76 | 5.51 | 6.41 | 3.84 | 5.52 | 5.7 | - | |
| Pocket size d) | Top1 | 22.2 | 7.1 | 50.0 | 0.0 | 0.0 | 0.0 | 26.7 | 100.0 | 0.0 | 22.8 | 32 |
| Top3 | 55.6 | 78.6 | 80.0 | 0.0 | 0.0 | 81.2 | 60.0 | 100.0 | 0.0 | 50.6 | 66 | |
| Avg. Distance | 0.20 | 0.13 | 0.15 | 0.30 | 0.47 | 0.11 | 0.10 | 0.06 | 0.30 | 0.20 | - | |
| Random retrieval e) | Top 1 | 9.8 (σ=10.1) | 13.2 (9.4) | 9.7 (9.2) | 6.3 (9.8) | 5.4 (9.5) | 15.4 (8.7) | 14.4 (9.0) | 19.4 (9.1) | 6.2 (11.2) | 11.0 | 15.2 |
| Top 3 | 28.0 (σ=14.7) | 39.7 (12.7) | 30.7 (14.5) | 20.8 (16.3) | 17.0 (16.4) | 45.0 (12.8) | 42.1 (13.0) | 54.5 (11.2) | 19.0 (18.0) | 33.0 | 34.9 |
The Euclidean distance is used for measuring similarity of pockets, and the Equation 7 is used for finally predicting bound ligands. The Top1 and Top3 success rate by random chance is shown in the parentheses in the rows of the geometry-based descriptor.
The geometry-based descriptor (G) denotes the pseudo-Zernike moments combined with the pocket size using an optimal weighting factor (Eqns. 12, 13). The electrostatic potential-based descriptor (E) uses the pseudo-Zernike moments of the electrostatic potential of the pocket surface combined with the pocket size using an optimal weighting factor. G+E uses optimally weighted Euclidean distance of G and E.
The average value of the success rate of the nine ligands.
The number of pockets which successfully retrieved pockets of the same type within Top1 or Top3.
The pocket size information only (Table 3) is used to retrieve pockets in the dataset.
The average and the standard deviation (in the parentheses) of 500 random trials are shown.