Kundrotas Petras J, Alexov Emil G
Computational Biophysics and Bioinformatics, Department of Physics, Clemson University, Clemson, SC 29634, USA.
BMC Bioinformatics. 2006 Nov 16;7:503. doi: 10.1186/1471-2105-7-503.
Predicting residues' contacts using primary amino acid sequence alone is an important task that can guide 3D structure modeling and can verify the quality of the predicted 3D structures. The correlated mutations (CM) method serves as the most promising approach and it has been used to predict amino acids pairs that are distant in the primary sequence but form contacts in the native 3D structure of homologous proteins.
Here we report a new implementation of the CM method with an added set of selection rules (filters). The parameters of the algorithm were optimized against fifteen high resolution crystal structures with optimization criterion that maximized the confidentiality of the predictions. The optimization resulted in a true positive ratio (TPR) of 0.08 for the CM without filters and a TPR of 0.14 for the CM with filters. The protocol was further benchmarked against 65 high resolution structures that were not included in the optimization test. The benchmarking resulted in a TPR of 0.07 for the CM without filters and to a TPR of 0.09 for the CM with filters.
Thus, the inclusion of selection rules resulted to an overall improvement of 30%. In addition, the pair-wise comparison of TPR for each protein without and with filters resulted in an average improvement of 1.7. The methodology was implemented into a web server http://www.ces.clemson.edu/compbio/recon that is freely available to the public. The purpose of this implementation is to provide the 3D structure predictors with a tool that can help with ranking alternative models by satisfying the largest number of predicted contacts, as well as it can provide a confidence score for contacts in cases where structure is known.
仅使用一级氨基酸序列预测残基间的接触是一项重要任务，它可指导三维结构建模并验证预测的三维结构的质量。相关突变（CM）方法是最具前景的方法，已被用于预测在一级序列中距离较远但在同源蛋白质的天然三维结构中形成接触的氨基酸对。
在此，我们报告了一种添加了一组选择规则（过滤器）的CM方法的新实现方式。针对15个高分辨率晶体结构对算法参数进行了优化，优化标准是使预测的可信度最大化。优化后，无过滤器的CM的真阳性率（TPR）为0.08，有过滤器的CM的TPR为0.14。该方案进一步以65个未包含在优化测试中的高分辨率结构为基准进行测试。基准测试结果显示，无过滤器的CM的TPR为0.07，有过滤器的CM的TPR为0.09。
因此，纳入选择规则使整体提升了30%。此外，对每种蛋白质有无过滤器时的TPR进行成对比较，平均提升了1.7。该方法已在一个网络服务器（http://www.ces.clemson.edu/compbio/recon）上实现，公众可免费使用。此实现的目的是为三维结构预测者提供一种工具，该工具可通过满足最多数量的预测接触来帮助对替代模型进行排名，并且在已知结构的情况下可为接触提供置信度得分。
