Protein tertiary structure is important information in various areas of biological research, however, the experimental cost associated with structure determination is high, and computational prediction methods have been developed to facilitate a more economical approach. Currently, template-based modeling methods are considered to be the most practical because the resulting predicted structures are often accurate, provided an appropriate template protein is available. During the first stage of template-based modeling, sensitive homology detection is essential for accurate structure prediction. However, sufficient structural models cannot always be obtained due to a lack of quality in the sequence alignment generated by a homology detection program. Therefore, an automated method that detects remote homologs accurately and generates appropriate alignments for accurate structure prediction is needed. In this paper, we propose an algorithm for suitable alignment generation using an intermediate sequence search for use with template-based modeling. We used intermediate sequence search for remote homology detection and intermediate sequences for alignment generation of remote homologs. We then evaluated the proposed method by comparing the sensitivity and selectivity of homology detection. Furthermore, based on the accuracy of the predicted structure model, we verify the accuracy of the alignments generated by our method. We demonstrate that our method generates more appropriate alignments for template-based modeling, especially for remote homologs. All source codes are available at https://github.com/shuichiro-makigaki/agora.

蛋白质三级结构在生物学研究的各个领域都是重要的信息，但是，与结构确定相关的实验成本很高，并且已经开发了计算预测方法来促进更经济的方法。当前，基于模板的建模方法被认为是最实用的，因为只要有合适的模板蛋白可用，所得到的预测结构通常是准确的。在基于模板的建模的第一阶段，敏感的同源性检测对于准确的结构预测至关重要。然而，由于同源性检测程序所产生的序列比对缺乏质量，不能总是获得足够的结构模型。因此，需要一种自动检测远程同源物并生成适当比对以进行准确结构预测的自动化方法。在本文中，我们提出了一种使用中间序列搜索进行适合的比对生成的算法，以用于基于模板的建模。我们使用中间序列搜索进行远程同源性检测，并使用中间序列进行远程同源物的比对生成。然后，我们通过比较同源性检测的灵敏度和选择性来评估所提出的方法。此外，基于预测结构模型的准确性，我们验证了由我们的方法生成的路线的准确性。我们证明，对于基于模板的建模，尤其是对于远程同源物，我们的方法可以生成更合适的比对。所有源代码都可以在https：// github上获得。