How good are predictions of protein secondary structure?

Abstract

The three most widely used methods for the prediction of protein secondary structure from the amino acid sequence are tested on 62 proteins of known structure using a program package and data collection not previously available. None of these methods predicts better than 56% of the residues correctly, for a three state model (helix, sheet and loop). The algorithms of Robson et al. [J. Mol. Biol. (1978) 120, 97-120] and Lim [J. Mol. Biol. (1974) 88, 873-894] are the best of those tested. New methods, now under development, can be tested against this benchmark.