Table 1.
Prediction performance in a blinded test on known structures
| PDB ID (no. strands) | Gene name | Amino acid count | No. sequences in alignment | No. sequences per amino acid | PPV of top L/2 ECs | PPV of top L/2 ECs between adjacent strands | rmsd of the top-ranked model† | TM score of the top-ranked model† |
| 1p4t (8) | NspA | 160 | 19,458 | 121.6 | 0.66 | 0.72 | 1.59 (87) | 0.85 |
| 1kmp (22) | FecA | 535 | 48,590 | 90.8 | 0.79 | 0.82 | 4.48 (217) | 0.67 |
| 3kvn (12) | EstA | 287 | 20,273 | 70.6 | 0.7 | 0.72 | 3.12 (124) | 0.68 |
| 2j1n (16) | OmpC | 353 | 18,764 | 53.2 | 0.64 | 0.77 | 4.65 (152) | 0.58 |
| 3ohn (24) | FimD | 534 | 15,557 | 29.1 | 0.59 | 0.6 | 4.77 (150) | 0.46 |
| 4k3c (16) | BamA | 384 | 10,014 | 26.1 | 0.65 | 0.67 | 5.22 (141) | 0.48 |
| 4e1t (12) | InvA | 251 | 5,778 | 23.0 | 0.42 | 0.57 | 4.49 (93) | 0.45 |
| 1t16 (14) | FadL | 387 | 5,422 | 14.0 | 0.63 | 0.69 | 3.86 (147) | 0.66 |
| 3syb (18) | OpdP | 430 | 5,031 | 11.7 | 0.58 | 0.65 | 4.15 (162) | 0.62 |
| 2wjr (12) | NanC | 223 | 2,552 | 11.4 | 0.5 | 0.66 | 4.38 (109) | 0.50 |
| 1thq* (8) | PagP | 161 | 1,836 | 11.4 | 0.34 | 0.55 | NA | NA |
| 1qd6 (12) | OmpLA | 250 | 2,367 | 9.5 | 0.46 | 0.58 | 3.5 (112) | 0.61 |
| 4q35 (26) | LptD | 585 | 5,285 | 9.0 | 0.44 | 0.62 | 6.54 (174) | 0.36 |
| 1a0s (18) | ScrY | 423 | 3,299 | 7.8 | 0.2 | 0.55 | 6.12 (139) | 0.39 |
| 1tly (12) | Tsx | 270 | 2,039 | 7.5 | 0.27 | 0.56 | 5.06 (103) | 0.41 |
| 2jk4 (19) | VDAC | 283 | 1,914 | 6.8 | 0.22 | 0.5 | 6.18 (157) | 0.45 |
| 2erv (8) | PagL | 159 | 1,006 | 6.3 | 0.25 | 0.48 | 2.55 (79) | 0.67 |
| 4gey* (16) | OprB | 420 | 2,376 | 5.7 | 0.47 | 0.56 | NA | NA |
| 2o4v (16) | OprP | 397 | 2,182 | 5.5 | 0.3 | 0.52 | 4.98 (109) | 0.40 |
Seventeen of the 19 proteins with a known structure are folded in a blinded test such that no known structural information is used for folding. Positive predictive value (PPV) is the ratio of correct predictions (shortest distance between two residues in the known structure ≤ 5 Å) and all predictions made. Three-dimensional models are compared with the known structure, using a TM score that ranges from 0 to 1 and an rmsd value. The closer the TM score is to 1, the more similar the model is to the known structure. rmsd values closer to 0 signify high structure similarity. rmsd values of residues aligned (shown in brackets) is reported in angstroms. PDB, Protein Data Bank.
PagP and OprB are not applicable (NA) cases for folding, as boctopus2 strand assignment is incorrect.
TM score and rmsd of the predicted transmembrane β-barrel region excluding the loops.