Ab initio protein structure prediction results using a simple distance geometry-based method

An approach to ab initio protein tertiary structure prediction based on building and scoring libraries of folds was tested at the CASP3 experiment. We present blind predictions for five helical targets varying in length from 31 to 114 residues. Our method was able to predict most of each structure correctly to about 6 Å Cα root-meansquare deviation (RMSD), with the exception of the largest target. Because distance geometry is used to constrain every pair of predicted helices to a generic distance range, inadequate sampling of tertiary folds occurs for larger targets or targets with poorly assigned helical residues. However, for targets with 5 helices or fewer, the native interhelical distances are more compatible with the generic bounds, and sampling improves to the extent that near-native fold selection is possible.