Prediction of the Number of Residue Contacts in Proteins

摘要

Knowing the number of residue contacts in a protein is crucial for deriving constraints useful in modeling protein folding and/or scoring remote homology search. Here we focus on the prediction of residue contacts and show that this figure can be predicted with a neural network based method. The accuracy of the prediction is 12 percentage points higher than that of a simple statistical method. The neural network is used to discriminate between tow different states of residue contacts, characterized by a contact number higher or lower than the average value of the residue distribution. When evolutionary information is taken into account, our method correctly predicts 695 of the residue states in the data base and it adds to the prediction of residue solvent accessibility. The predictor is available at htpp://www.biocomp.unibo.it