Table 1. Average model quality scores for different variations of template selection strategies and restraints used with Modeller on a test set of 1000 single- and multi-domain proteins in the pdb20 database.
The GDC-all score is similar to GDT-ha but also includes side-chain atoms in its assessment. Percent improvements are with respect to the first line. P-values are calculated based on a paired t-test with respect to the GDT-ha score in the previous line.
| Method | GDT-ha | P-value | GDC-all | |||
|---|---|---|---|---|---|---|
| Name | Templates | Template selection | Restraints | |||
| s.1st.old | Single | first hit | Modeller | 0.443 (+0%) | - | 51.9 (+0%) |
| s.NN.old | Single | neural network | Modeller | 0.447 (+0.9%) | 1.5E-6 | 52.4 (+1.0%) |
| s.NN.new | Single | neural network | new | 0.450 (+1.5%) | 8E-6 | 52.8 (+1.7%) |
| m.ss.old | Multiple | simple selection | Modeller | 0.462 (+4.3%) | 1E-10 | 53.5 (+3.1%) |
| m.mt.old | Multiple | new multi-template | Modeller | 0.480 (+8.4%) | 2E-16 | 55.1 (+6.2%) |
| m.mt.new | Multiple | new multi-template | new | 0.492 (+11.1%) | 2E-16 | 56.3 (+8.5%) |