14种结合自由能评价函数的比较

采用LigandFit作为构象采样工具，以230个蛋白质-配体复合物组成的预测集，系统地比较了14种自由能评价函数（Ligscore1、Ligscore2、Plp1、Plp2、Jain、Pmf、Ludi1、Ludi2、Ludi3、D-score、Pmf-score、G-score、Chemscore以及Xscore）对蛋白质和小分子之间的结合模式以及结合自由能的预测能力. Plp1、Plp2、G-score、Pmf和Xscore在预测测试集结合自由能时得到的分数同实验测定的结合自由能的线性相关系数大于50%. 在识别配体分子实验结合构象的能力方面， 选择测试构象与实际构象间的位置均方根偏差rmsd≤0.20 nm作为评价标准，14种评价函数的成功率从46%到77%不等，其中Ligscore1、Ligscore2、Plp1、Plp2以及Xscore的成功率都在70%以上. 将评价函数中的2个或者3个组合得到一组共同评价函数可以进一步提高实验构象的预测能力， 其预测成功率可以达到80%. 实验表明Xscore、Plp1和Plp2在对接和评价方面都得到较好的结果.

Comparative Studies of 14 Binding Free Energies Scoring Functions

Fourteen scoring functions (Ligscore1, Ligscore2, Plp1, Plp2, Jain, Pmf, Ludi1, Ludi2, Ludi3, D-score, Pmf-score, G-score, Chemscore and Xscore) have been compared on a testing set containing 230 protein-ligand complexes to evaluate their abilities to reproduce experimentally determined structures and binding free energies after conformational sampling by the program Ligandfit. When applied to predict the binding free energy of the 230 protein-ligand complexes, Plp1, Plp2, G-score, Pmf and Xscore can give linear correlation coefficients over 0.50 between experimental determined and computed values. When employed to identify the experimentally observed conformation, if using root-mean-square deviation(rmsd)≤0.20 nm as the criterion, Ligscore1, Ligscore2, Plp1, Plp2 and Xscore can gain success rate larger than 70%. Combining any two or three of the 14 scoring functions into a consensus scoring function further improves the success rate to 80%. The results suggest that Xscore, plp1 and plp2 are more practicable than others in docking and scoring.