Table 2. Comparison of Different Methods on Validation Datasets.
| accuracy
for DIBS dataset (698 complexes of IDP
binding to ordered proteins) |
accuracy
for MFIB dataset (157 complexes of IDP
binding to IDPs) |
accuracy for fully ordered dataset (543 complexes) | |||||
|---|---|---|---|---|---|---|---|
| methods | disorder prediction on IDPs (%) | order prediction on IDP partners (%) | complex predictionc (%) | disorder prediction on IDPs (%) | disorder prediction on IDP partners (%) | complex prediction (%) | order prediction (%) |
| NMA–NIA | 99.7 | 92.1 | 91.8 | 84.1 | 77.7 | 77.1 | 91.6 |
| Rg/N | 99.1 | 96.8 | 93.7 | 34.4 | 46.6 | 29.3 | 98.8 |
| CDFb | 70.1 | 83.4 | 58.1 | 42.9 | 41.1 | 35.3 | 89.2 |
| CH_value | 67.5 | 84.8 | 57.3 | 34.0 | 37.4 | 28.2 | 95.6 |
a
Bold and underline highlights indicate the best prediction accuracy by different methods for different datasets in the columns.
b
CDF predictor does not have results for short chains (<30 amino acids). The percentage showing here only includes chains that are longer than 30 amino acids.
c
Complex prediction means that both components in the protein complexes are correctly predicted.