Improved ab initio predictions with a simplified, flexible geometry model.

Structure predictions for nine targets from the CASP3 meeting are presented and compared with the experimental structure. These predictions are made using the simplified flexible geometry representation of protein structure and potentials which mimic the physical forces involved in protein folding with no help from multiple sequences. The major differences from the CASP2 potentials are identified, and the prediction successes and failures are related to the underlying potentials. Target T0065 was successfully folded from an extended chain to 3.8 A CA RMS of the native structure. The quality of the secondary structure component of all predictions was significantly improved, averaging a Q3 of 64%, which was spread evenly across the fold types, all alpha, alpha/beta mixed, and mainly beta. A number of the other predictions had a spatial arrangement of the secondary structure segments close to native, although the major problem with most predictions was the over-extension of secondary structure segments which often coalesced independent segments together. For target T0056 the ab initio prediction was the closest to the native structure of all methods including threading according to the Hubbard RMS coverage plots.