Department of Health Statistics, Second Military Medical University, Shanghai 200433, China.
Biochem Biophys Res Commun. 2013 Sep 20;439(2):303-8. doi: 10.1016/j.bbrc.2013.08.023. Epub 2013 Aug 20.
G-protein-coupled receptors (GPCRs) constitute a remarkable protein family of receptors that are involved in a broad range of biological processes. A large number of clinically used drugs elicit their biological effect via a GPCR. Thus, developing a reliable computational method for predicting the functional roles of GPCRs would be very useful in the pharmaceutical industry. Nowadays, researchers are more interested in functional roles of GPCRs at the finest subtype level. However, with the accumulation of many new protein sequences, none of the existing methods can completely classify these GPCRs to their finest subtype level. In this paper, a pioneer work was performed trying to resolve this problem by using a hierarchical classification method. The first level determines whether a query protein is a GPCR or a non-GPCR. If it is considered as a GPCR, it will be finally classified to its finest subtype level. GPCRs are characterized by 170 sequence-derived features encapsulating both amino acid composition and physicochemical features of proteins, and support vector machines are used as the classification engine. To test the performance of the present method, a non-redundant dataset was built which are organized at seven levels and covers more functional classes of GPCRs than existing datasets. The number of protein sequences in each level is 5956, 2978, 8079, 8680, 6477, 1580 and 214, respectively. By 5-fold cross-validation test, the overall accuracy of 99.56%, 93.96%, 82.81%, 85.93%, 94.1%, 95.38% and 92.06% were observed at each level. When compared with some previous methods, the present method achieved a consistently higher overall accuracy. The results demonstrate the power and effectiveness of the proposed method to accomplish the classification of GPCRs to the finest subtype level.
G 蛋白偶联受体（GPCRs）是一类重要的受体家族，参与广泛的生物学过程。大量的临床应用药物通过 GPCR 发挥其生物学效应。因此，开发一种可靠的计算方法来预测 GPCR 的功能作用在制药工业中非常有用。如今，研究人员对 GPCR 最精细亚型水平的功能作用更感兴趣。然而，随着许多新蛋白质序列的积累，现有的方法都不能将这些 GPCR 完全分类到其最精细的亚型水平。在本文中，通过使用层次分类方法，进行了一项开创性的工作来试图解决这个问题。第一级确定查询蛋白质是否为 GPCR 或非 GPCR。如果它被认为是 GPCR，它将最终被分类到其最精细的亚型水平。GPCR 的特征是 170 个序列衍生特征，这些特征既包含氨基酸组成又包含蛋白质的理化特征，并且支持向量机被用作分类引擎。为了测试本方法的性能，构建了一个非冗余数据集，该数据集按照七个层次组织，涵盖了比现有数据集更多的 GPCR 功能类别。每个层次的蛋白质序列数量分别为 5956、2978、8079、8680、6477、1580 和 214。通过 5 倍交叉验证测试，在每个层次观察到的总体准确率分别为 99.56%、93.96%、82.81%、85.93%、94.1%、95.38%和 92.06%。与一些先前的方法相比，本方法始终获得更高的总体准确率。结果表明，该方法在将 GPCR 分类到最精细的亚型水平方面具有强大的功效和有效性。
