Department of Biochemistry and Bioinformatics Centre, Bose Institute, P-1/12 CIT Scheme VIIM, Kolkata 700 054, India.
J Chem Inf Model. 2010 Dec 27;50(12):2266-73. doi: 10.1021/ci1002922. Epub 2010 Nov 16.
Computational tools are available today for the detection and delineation of the clefts and cavities in protein 3D structure and ranking them on the basis of probable binding site clefts. There is a need to improve the ranking of clefts and accuracy of predicting catalytic site clefts. Our results show that the distance of the clefts from protein centroid and sequence entropy of the lining residues, when used in conjunction with the volume, are valuable descriptors for predicting the catalytic site. We have applied the SVM approach for recognizing and ranking the active site clefts and tested its performance using different combinations of attributes. In both the ligand-bound and the unbound forms of structures, our method correctly predicts the active site clefts in 73% of cases at rank one. If we consider the results at rank 3 (i.e., the correct solution is among one of the top three solutions), the correctly predicted cases are 94% and 90% for the bound and the unbound forms of structures, respectively. Our approach improves the ranking of binding site clefts in comparison with CASTp and is comparable to other existing methods like Fpocket. Although the data set for training the SVM approach is rather small in size, the results are encouraging for the method to be used as complementary to other existing tools.
目前有一些计算工具可用于检测和描绘蛋白质三维结构中的裂隙，并根据可能的结合位点裂隙对其进行排序。需要改进裂隙的排序和催化位点预测的准确性。我们的研究结果表明，裂隙与蛋白质质心的距离以及衬砌残基的序列熵与体积结合使用，是预测催化位点的有价值描述符。我们已经应用 SVM 方法来识别和对活性位点裂隙进行排序，并使用不同的属性组合来测试其性能。在配体结合和未结合结构的形式中，我们的方法在排名第一的情况下正确预测了 73%的活性位点裂隙。如果我们考虑排名 3 的结果（即正确的解决方案在排名前三的解决方案之一中），则对于结合和未结合结构形式，正确预测的情况分别为 94%和 90%。与 CASTp 相比，我们的方法提高了结合位点裂隙的排序，并且与 Fpocket 等其他现有方法相当。尽管用于训练 SVM 方法的数据集规模较小，但对于该方法作为其他现有工具的补充，结果令人鼓舞。
