基于结构比对模板库的蛋白质模型评估

Evaluating Models Predicted Based on Structure Alignment Template Library

蛋白质天然构象预测是计算生物学领域最具有挑战性的课题之一。基于模板的预测方法是目前最为准确的方法,该方法的预测模型的好坏很大程度上在于其模板质量的好坏。从SCOP数据库中筛选了3 867个蛋白质,通过结构比对和统计分析,建立了一个基于结构比对的模板库;接着,利用动态归一化法和Profile-profile的原理,分别编写的搜索和比对程序;最后利用MODELLER的建模程序给出了蛋白质的三级结构模型。测试集由48个蛋白组成,首先,用Profile-profile搜索基于结构比对模板库获得同源模板,以此模板模建出蛋白质三级结构模型,同时用MODELLER中的搜索程序搜索仅有序列构成的序列库,最后同样获得蛋白质的三级模型。从测试结果的正确率看,该方法较MODELLER有了14.59%的提高;通过模型评估,可以看出该方法预测出的模型质量整体上也优于原有的MODELLER方法。因此,认为用profile-profile搜索基于结构比对模板库的方法要优于MODELLER中的搜索的方法。

Protein tertiary structure prediction is one of the most challenging topics in computational biology. The most accurate way is template-based method,in which the template is very important,and it determines the quality of the predicted models. In our study,3 867 proteins are selected from SCOP database,through structure alignment and statistics,a new template library based on structure alignment has been built then,by the method dynamic normalization and profile-profile,we write search and alignment program by C + +; finally,the predicted models were obtained by a program of get-model of MODELLER_9. 12. The test set consists of 48 proteins firstly,we use profile-profile method to search the homology,and the final model is built based on the template; at the same time,the target sequence is also predicted by MODELLER program. From the result,the accurate rate of model built by homology template obtained by the method of profile-profile has improved 14. 59% compared with MODELLER method,by evaluating the predicted models,the quality of the models is also superior than that obtained with MODELLER. So,a conclusion can be drawn that the method profile-profile to search structure alignment template library is superior to the MODELLER method.