Protein structure prediction from predicted residue properties utilizing a digital encoding algorithm.

Although many disparate methods have been applied to the problem, the accuracy of protein structural prediction still remains disappointingly low, averaging about 65% correct secondary structure assignment. A novel predictive method is presented here, which attempts to address some of the shortfalls inherent in representing a protein as a simple text-like sequence of amino acids, by deriving pattern-matching data from the predicted physical properties of a protein chain rather than from the sequence itself. A unique binary encoding algorithm is used to enable the property profiles to be correlated with known secondary structure, and hence to predict secondary structures for proteins with unknown structures. By treating the sequence in this manner, predictive accuracies averaging over 75% have been achieved.