Protein Structure Prediction Based On Fragment Database

The protein plays an important role in the organism. The protein function depends on the protein structure and the protein structure rests with the protein sequence. How to predict the protein structure according to their sequence is a fascinating yet unsolved problem, and also is the focus of our research.Usually, there are several methods to predict the protein structure as follows: If the target sequence is quite similar to a protein whose structure we have known, Homology Modeling is the best choice; If the target sequence has a poor similarity to any one protein, Threading is suitable; If neither homologous proteins nor remote homologous proteins exist, AB Initio works.Protein structure prediction based on fragment database belongs to ab initio, and in the Sixth Critical Assessment in Structure Prediction, it is combined with homology modeling: In the tactic region , the coordinates are from the template, and in the non-tactic region, the coordinates is simulated in the method. Before, we predict with the fragments of the same length. Though we can get a fairish result, the local optimal result can't necessarily lead to the global optimal result. It makes local prediction more difficult. So we try to reveal the relationship between the local sequence and the global sequence more accurately, with fragments that have definite structures and are quite conservative in the sequence.We propose a new prediction method based on fragment database so as to get the structure of the target sequence with information as little as possible. First, the proteins whose structures are known are split into fragments according to their structures, and then the fragments are clustered to form a fragment database. The fragments mentioned above are a new level of description to well understand and analyze the organization of protein structures. Second, the target sequence is cut into segments and each segment is assigned a certain structure according to the sequence information and the local optimal match algorithm. Then we use the shortest path algorithm to select optimal combinations to"cover"the protein sequence. The input is a target protein sequence and a library of fragments created by clustering fragments that were generated by cutting all protein structures. The output is a set of cutout fragments. The method can simulate the target sequence more accurately and reduce the computational complexity dramatically.