Table 16 Long range accuracy evaluation 5 with alternative feature selection. Average long range accuracy values at δ = 2 resolution for the top 10, top L/10, and top L/5 contact predictions from various predictors (with our methods listed in bold font) evaluated on CASP11 targets and sorted by Top 10 accuracy scores. These models were trained with features selected according to the mean decrease accuracy metric instead of mean decrease Gini
From: Predicting protein residue-residue contacts using random forests and deep networks
Method | Top 10 | L/10 | L/5 |
|---|---|---|---|
CONSIP2 (94) | 71.71% | 70.18% | 67.30% |
RaptorX-Contact (94) | 61.60% | 56.80% | 55.63% |
svm (78) | 56.67% | 54.62% | 50.60% |
MLiD (94) | 53.51% | 49.98% | 46.99% |
rf_select (78) | 52.44% | 52.30% | 51.09% |
DCA _cpp(88) | 47.07% | 44.10% | 40.82% |
sda_balanced (78) | 46.79% | 46.16% | 43.76% |
sda_unbalanced (78) | 40.14% | 38.25% | 37.12% |
IASL-COPE (91) | 37.58% | 34.87% | 33.70% |
MULTICOM-NOVEL (94) | 27.97% | 27.25% | 27.12% |
raghavagps-paaint (76) | 26.50% | 26.37% | 26.68% |
Pcons-net (87) | 17.72% | 18.41% | 18.93% |
FoDTcm (32) | 17.71% | 17.45% | 18.10% |
PLCT (91) | 16.87% | 17.24% | 17.08% |