Table 1.
Correlation coefficient and Z-score for different input parameters to the neural networks
| Input parameters | Predicting LGscore correlation/Z-score | Predicting MaxSub correlation/Z-score |
| Atom-3 contacts | 0.41/0.9 | 0.30/0.9 |
| Atom-13 contacts | 0.48/1.1 | 0.32/0.9 |
| Residue-6 contacts | 0.43/1.1 | 0.31/0.9 |
| Residue-20 contacts | 0.35/0.8 | 0.25/0.6 |
| Surface accessibility <25% | 0.51/1.4 | 0.38/1.2 |
| Surface accessibility 25%–50% | 0.21/0.4 | 0.06/0.2 |
| Surface accessibility all | 0.53/1.5 | 0.46/1.3 |
| Atom-13 + Residue-6 | 0.53/1.4 | 0.39/1.1 |
| Atom-13 + Residue-6 + Surface all | 0.63/2.0 | 0.50/1.5 |
| Atom-13 + Residue-6 + Surface all + Q3 | 0.71/2.4 | 0.58/2.1 |
| Atom-13 + Residue-6 + Surface all + Q3 + Cα | 0.74/2.6 | 0.60/2.2 |
| Atom-13 + Residue-6 + Surface all + Q3 + Cα + fatness | 0.75/2.7 | 0.62/2.4 |
| Atom-13 + Residue-6 + Surface all + Q3 + Cα + fatness + frac | 0.76/2.7 | 0.71/2.6 |
Atom-3 is the atom contacts between three different atom types, Atom-13 the contacts between 13 different atom types, Residue-6 and Residue-20 the contacts between 6 and 20 different residue types, respectively. Q3 is the fraction of similarity between predicted secondary structure and the secondary structure in the model. Cα is the difference between the all-atom model and the aligned Cα coordinates from the template that were used to build the model, as measured by LGscore and MaxSub for networks predicting LGscore and MaxSub, respectively. frac is the fraction of the protein that is modeled.