Table 6. Comparative performances of existing well-known glycosylation prediction tools and GlycoPP models on independent dataset of prokaryotic glycoproteins.
| Prediction of N-glycosites | ||||||
| Models (Threshold) | NetNglyc1 (0.5) | EnsembleGly3 (0.7) | GlycoPP-BPP (−0.1) | GlycoPP-CPP (0.3) | GlycoPP-PPP (−0.2) | GlycoPP-BPP+ASA |
| Sensitivity (%) | 88.89 | 94.44 | 89.47 | 68.42 | 78.95 | 89.47 |
| Specificity (%) | 25.00 | 11.36 | 73.68 | 73.68 | 73.68 | 84.21 |
| Accuracy (%) | 43.55 | 35.48 | 81.58 | 71.05 | 76.32 | 86.84 |
| MCC (%) | 0.15 | 0.09 | 0.64 | 0.42 | 0.53 | 0.74 |
| Prediction of O-glycosites | ||||||
| Models | NetOGlyc2 (0.1) | EnsembleGly3 (0.3) | GlycoPP-BPP (0.2) | GlycoPP-CPP (0.2) | GlycoPP-PPP (0) | GlycoPP-PPP+ASA |
| Sensitivity (%) | 100.00 | 6.67 | 72.13 | 72.55 | 77.05 | 81.97 |
| Specificity (%) | 3.19 | 93.05 | 73.77 | 68.18 | 70.49 | 70.49 |
| Accuracy (%) | 8.27 | 88.28 | 72.95 | 70.36 | 73.77 | 76.23 |
| MCC (%) | 0.04 | −0.00 | 0.46 | 0.41 | 0.48 | 0.53 |
Footnotes: 1: http://www.cbs.dtu.dk/services/NetNGlyc/, 2: http://www.cbs.dtu.dk/services/NetOGlyc-3.0/, 3: http://turing.cs.iastate.edu/EnsembleGly/, BPP- Binary profile of patterns, CPP- Composition profile of patterns, PPP- PSSM profile of patterns, MCC- Matthews correlation coefficient, AUC- Area under curve, SS-secondary structure and ASA- Accessible surface area.