Abstract: Novel and more efficient structural prediction methods are needed in order to reduce the everincreasing gap between the number of known protein sequences and structures. Better predictions of non-local contacts between residues could help improve fold recognition and de novo modeling. In the present study, we introduce PROFcon, a novel contact prediction method based on a neural network that combines information from various sources. PROFcon predictions are then used to estimate experimental folding rates of two-state proteins (i.e. proteins that fold without intermediate states).