Table 4. Performance of conversion to all-atom structures from Cα-traces of simulation snapshots.
See also Figure S7.
| Input | Method | RMSD1 | χ-angle accuracy1,2 | MolProbity1 | |||||
|---|---|---|---|---|---|---|---|---|---|
|
| |||||||||
| Backbone [A] | Heavy atom [A] | χ1 [%] | χ1+2 [%] | Score | Clash score | Rama favor [%] | Rotamer outlier [%] | ||
|
| |||||||||
| Cα from all-atom MD snapshots | MD snapshots | 1.45 (0.23) | 0.8 (0.5) | 93.5 (2.1) | 2.7 (1.2) | ||||
|
| |||||||||
| cg2all | 0.25 (0.04) | 1.18 (0.12) | 77.7 (3.4) | 58.7 (4.6) | 2.31 (0.23) | 37.1 (9.8) | 96.4 (1.5) | 1.0 (0.7) | |
| w/SCRWL3 | 1.27 (0.14) | 76.2 (3.6) | 58.1 (4.8) | 2.20 (0.19) | 32.5 (8.0) | 96.4 (1.5) | 0.0 (0.1) | ||
| cg2all (fixed)4 | 0.25 (0.04) | 1.15 (0.12) | 77.9 (3.4) | 58.8 (4.6) | 2.42 (0.24) | 41.0 (10.1) | 96.3 (1.5) | 1.4 (0.8) | |
|
| |||||||||
| PULCHRA5 | 0.52 (0.16) | 1.70 (0.14) | 54.2 (4.0) | 32.1 (3.9) | 3.84 (0.20) | 162.7 (25.5) | 85.1 (3.8) | 5.2 (1.6) | |
| w/SCWRL3 | 1.51 (0.15) | 68.0 (4.2) | 49.7 (4.8) | 2.93 (0.17) | 67.1 (15.5) | 85.1 (3.8) | 0.1 (0.2) | ||
|
| |||||||||
| REMO5,6 | 0.94 (0.47) | 2.21 (0.46) | 44.5 (4.5) | 26.4 (4.8) | 4.42 (0.21) | 197.5 (37.2) | 75.1 (7.9) | 15.6 (3.9) | |
| w/SCWRL3 | 1.95 (0.53) | 63.5 (5.9) | 45.5 (6.5) | 3.22 (0.24) | 102.4 (45.0) | 75.1 (7.9) | 0.2 (0.3) | ||
|
| |||||||||
| Cα after minimization with COCOMO7 | cg2all | 0.43 (0.13) | 1.34 (0.19) | 73.7 (4.2) | 54.8 (5.1) | 2.55 (0.23) | 44.5 (12.5) | 94.4 (1.9) | 1.1 (0.8) |
| w/SCWRL3 | 1.43 (0.19) | 72.3 (4.1) | 53.5 (5.2) | 2.38 (0.18) | 36.1 (10.2) | 94.4 (1.9) | 0.1 (0.1) | ||
| cg2all (fixed)4 | 0.42 (0.13) | 1.32 (0.19) | 73.3 (4.3) | 54.3 (5.0) | 2.70 (0.24) | 50.1 (13.3) | 94.1 (2.0) | 1.7 (0.9) | |
|
| |||||||||
| PULCHRA5 | 0.61 (0.18) | 1.79 (0.18) | 52.5 (4.1) | 31.0 (4.1) | 3.87 (0.19) | 156.8 (25.3) | 83.5 (3.9) | 5.6 (1.7) | |
| w/SCWRL3 | 1.62 (0.19) | 65.8 (4.4) | 47.1 (5.1) | 2.95 (0.15) | 65.1 (15.3) | 83.5 (3.9) | 0.1 (0.2) | ||
|
| |||||||||
| REMO5,6 | 1.16 (0.49) | 2.40 (0.53) | 43.3 (4.7) | 25.7 (4.8) | 4.50 (0.22) | 194.5 (40.1) | 69.9 (9.5) | 17.4 (4.3) | |
| w/SCWRL3 | 2.23 (0.59) | 59.4 (6.1) | 41.3 (6.6) | 3.33 (0.25) | 116.1 (48.2) | 69.9 (9.5) | 0.2 (0.3) | ||
The average reconstruction accuracy measures for the test set protein structures (n=720) are given with their standard deviations in the parentheses.
Side chain χ-angles were considered accurate when deviations from experimental values were less than 30 degrees.
Side chains were reconstructed using SCWRL4 after building a backbone structure using other methods (e.g., cg2all, PULCHRA, and REMO).
The atomic coordinates in the input files were preserved, while the original method does not. For instance, cg2all model for “Cα (fixed)” generates output structures with the exact same Cα coordinates of the input structures. On the other hand, the original cg2all model generates slightly altered Cα coordinates.
Multi-chain targets were converted chain-by-chain to all-atom structures and superposed onto the original Cα-trace.
Conversions of several structures failed because short peptides could not be handled.
All-atom MD simulation snapshots were considered as the ground truth. The average structure change in Cα-RMSD after minimization using COCOMO model was 0.30 Å.