Prediction of protein folding using residue fragment graphs

Understanding the folding process of proteins is among the most challenging problems (protein folding problem) for scientists in biochemistry, mathematics, physics and all other related sciences. Given a sequence of amino acids, they want to know how the interatomic forces between them yields a folded state, a specific three dimensional configuration in space which is stable. It has been shown that Alzheimer's and Mad Cow diseases are the result of misfolding that happens in some proteins. Discovering the mystery of protein folding will help in understanding the mechanism of infectious diseases and designing drugs with specific therapeutic properties.; In our work, we try to find the configuration for the C _α atoms of the amino acids in the target protein's chain. First we split the chain to the fragments of 7 residues long (7mers). Then we use the 7mer library of known proteins' C_α configurations created by J. Moult and H. Zhang to create a 7mer graph. We then try several methods to find a path through the 7mer graph which yields a good configuration. The constrained random path with side conditions (CRPSC) is the most successful one. In each trial of this method, we construct a random path so that the nodes on this path are within some RMSD cutoff of their neighbors. Configurations that do not pass the side conditions (compactness, bond length constraints) are ignored. The predicted configurations follow the pattern of the true configuration but are not correct in details. So using local information is not sufficient to precisely predict the global configuration of an unknown protein.; Most of the scientists that work in this field use root mean square distance (RMSD) between the true and predicted configuration to measure the accuracy of the prediction. They believe smaller RMSD means better prediction. Our result shows that this is not always the case. So we conclude using just RMSD is not a good way to evaluate a prediction method.