Protein fold recognition with a disulfide -cognizant scoring function.

Abstract

Protein structural knowledge is essential to understanding cellular function and the molecular basis of disease. Computational methods which can predict the correct protein structure for a given amino acid sequence would be a tremendous aid in biomedical and agricultural research. Homology modeling has proven useful but is precluded if the sequence of interest has low similarity to proteins of known structure. In these cases, fold recognition methods are more promising because they consider sequence to fold compatibility versus sequence to sequence similarity. Required for the success of fold recognition are a representative fold library and an accurate scoring function that assesses sequence-fold compatibility. Fold libraries are usually constructed using simplified representations of known proteins, with each side chain reduced to a single point in space. This simplification incurs uncertainty in the scoring function. Here, a scoring function is constructed to better quantify residue environment by providing accurate estimates of intramolecular contact despite the loss of detail in simplified models. Contact curves are derived and provide the average intramolecular interaction between amino acids given the distance between their simplified side chain representations. The improved quantification of residue environment is then used in the scoring function. Small, disulfide-bearing proteins have been particularly problematic in fold recognition tests. Previously, explicit consideration of disulfides has not been included in scoring functions. In this work, a disulfide recognition algorithm is developed to allow identification of disulfide bonds in simplified fold models. Disulfide bond characteristics are estimated accurately and provide a qualitative assessment of putative disulfide bonds. Each identified disulfide provides a favorable contribution to the scoring function based on bond quality. The scoring function, with little training and few adjustable parameters, is greater than 80% accurate in ungapped threading tests using a variety of proteins. The same level of accuracy is observed in a threading test of small proteins that have disulfide bonds. Fold recognition ability is greatly improved for disulfide-bearing proteins, yet does not detract from the predictive accuracy with the general population of proteins.