ACCURACY OF SIDE-CHAIN PREDICTION UPON NEAR-NATIVE PROTEIN BACKBONES GENERATED BY AB-INITIO FOLDING METHODS

The ab initio folding problem can be divided into two sequential tasks of approximately equal computational complexity: the generation of na tive-like backbone folds and the positioning of side chains upon these backbones. The prediction of side-chain conformation in this context is challenging, because at best only the near-native global fold of th e protein is known. To test the effect of displacements in the protein backbones on side-chain prediction for folds generated ab initio, set s of near-native backbones (less than or equal to 4 Angstrom C alpha R MS error) for four small proteins were generated by two methods. The s teric environment surrounding each residue was probed by placing the s ide chains in the native conformation on each of these decoys, followe d by torsion-space optimization to remove steric clashes on a rigid ba ckbone, We observe that on average 40% of the chi 1 angles were displa ced by 40 degrees or more, effectively setting the limits in accuracy for sidechain modeling under these conditions. Three different algorit hms were subsequently used for prediction of side-chain conformation. The average prediction accuracy for the three methods was remarkably s imilar: 49% to 51% of the chi 1 angles were predicted correctly overal l (33% to 36% of the chi 1+2 angles). Interestingly, when the inter-si de-chain interactions were disregarded, the mean accuracy increased. A consensus approach is described, in which side-chain conformations ar e defined based on the most frequently predicted chi angles for a give n method upon each set of near-native backbones, We find that consensu s modeling, which de facto includes backbone flexibility, improves sid e-chain prediction: chi 1 accuracy improved to 51-54% (36-42% of chi 1 +2). Implications of a consensus method for ab initio protein structur e prediction are discussed. (C) 1998 Wiley-Liss, Inc.