Abstract
While more and more completely sequenced genomes are available, the next important work is to know the three-dimension structures of the annotated sequences. However, large-scale determination of protein structures by X-ray crystallography and nuclear magnetic resonance spectroscopy is time-consuming and unpractical. Computing methods can speed up the structures determination. Homology modeling is the best method to predict the structures among the others. Since previous studies focused only on the relation between the local sequences alignments and the RMSD (root meaning square deviation) of the □ carbon atom, we want to focus on the relation between the global sequences alignments and the RMSD of the other atoms including all atom, backbone and side-chain. First, we got the samples from the PDB (Protein Data Bank, http://www.rcsb.org/pdb/). Second, we did the pair-wise global sequences alignments and predicted the three-dimension structures via homology modeling. Finally, we calculated the RMSD between the predicted structures and the original structures. The results indicate that better predicted structures would be obtained while the template sequences are longer than the target sequences. If the identity between the sequences is low, one could select related proteins with the similar functions or belonged to the same family to be your templates. When the identity is > 60%, the RMSD between the templates and the target would be almost the same. Besides, the cut-offs of least identity in order to get the fine structures, with RMSD below 3.5Å, for different atoms are 38~45%. Furthermore, we predicted structures for the whole genome of Xanthomonas campestris pv. campestris. As a result, we got 86 reasonable structures from 1291 annotated sequences.