Table III. Decoy Quality Comparison Between the HMM Model17 and CRFSampler.
| Test proteins | HMMa | CRFSamplerb | CRFSamplerc | |||||
|---|---|---|---|---|---|---|---|---|
| Name, PDB code | L | α,β | Good (%) | Best (Å) | Good (%) | Best (Å) | Good (%) | Best (Å) |
| Protein A, 1FC2 | 43 | 2,0 | 17.1 | 2.6 | 26.8 | 2.13 | 49.1 | 1.94 |
| Homeodomain, 1ENH | 54 | 2,0 | 12.2 | 3.8 | 16.7 | 2.29 | 22.4 | 2.32 |
| Protein G, 2GB1 | 56 | 1,4 | 0.0 | 5.9 | 26.4 | 3.05 | 23.3 | 2.91 |
| Cro repressor, 2CRO | 65 | 5,0 | 1.1 | 4.1 | 18.3 | 2.76 | 16.8 | 2.79 |
| Protein L7/L12, 1CTF | 68 | 3,1 | 0.35 | 4.1 | 3.0 | 4.04 | 2.4 | 3.70 |
| Calbidin, 4ICB | 76 | 4,0 | 0.38 | 4.5 | 0.24 | 4.45 | 0.51 | 4.63 |
Both sequence and secondary structure information are used. In total, 100,000 decoys are generated by the HMM model while only 20,000 decoys by each CRFSampler.
Columns 1–3 list name and PDB code, length and number of α-helices and β-strands of the test proteins. Column “Good” lists the percentage of good decoys (with RMSD ≤ 6 Å). Column “Best” lists the RMSD of the best decoy for each test protein.
Trained using true secondary structure and primary sequence while tested using predicted secondary structure (by PSIPRED38) and primary sequence and the results are taken from Ref. 17.
Trained and tested using predicted secondary structure (by PSIPRED) and primary sequence.
Trained and tested using predicted secondary structure likelihood scores (by PSIPRED) and PSI-BLAST sequence profile.