The Protein Structure Prediction Based On Distance Constraints

Protein plays an important role in organisms’ life and its structure is of great importance to the biology and medicine. Thus, predicting protein structures with computational methods is critical in bioinformatics. This thesis studies to mine distance constraints from the primary sequence of protein, and apply the mined distance constraints to the protein structure prediction.Protein structure is determined by the interactions among residues. There are a lot of distance constraints between residues which contains signi?cant information about protein structures and their functions.Evolutionary Coupling( EC) can be learned distance constraints by machine learning method. The coarse grained EC information is transformed into ?ne grained distance constraints between residues, which provides comparatively precise constraints for the protein structure prediction.A new sampling method for protein structure prediction is then proposed. The distance constraints are applied to sample the coordinate of decoys by improving the Hybrid Monte Carlo algorithm. When the distance constraints are added to sampling process, the accuracy of sampling is improved and we can obtain more decoys which is near-native structure.