Separating true positive predicted residue contacts from false positive ones in mainly α proteins, using constrained Metropolis MC simulations