The determination of a protein’s structure provides important in- formation that can be used for various practical applications in the biological sciences, such as virtual screening, function prediction, etc. Protein structures can be precisely predicted using template- based modeling if we can nd good template structures from a database. However, such predictions sometimes fail even if a tem- plate with su cient quality is found because the sequence align- ment used for the modeling is incorrect.
In this paper, we propose a new method for improving sequence alignment in single-template-based modeling. The sequence align- ments used as an input of template-based modeling are normally generated by homology search tools, and the alignments vary de- pending on the search algorithm used. Each single alignment is of- ten imperfect, but most of them have suitable parts for template- based modeling at di erent positions. Thus, a pro le of multiple alignments is typically constructed to obtain a consensus among the alignments by multiple template search tools. Integrated align- ments are generated by random sampling, and the nal prediction model is selected based on model quality assessment scores and the joint probability of the pro le.
We performed evaluation tests using template-based modeling targets in CASP11 and compared the proposed method to several existing major alignment algorithms. The results showed that the proposed method could improve the model accuracy of single-template modeling.
各種検索
サポート
ヘルプ