Comparing the quality and predictiveness between 3D QSAR models obtained from manual and automated alignment

A set of 113 flexible cyclic urea inhibitors of human immunodeficiency virus protease (HIV-1 PR) was used to compare the quality and predictive power of CoMFA and CoMSIA models for manually or automatically aligned inhibitor set. Inhibitors that were aligned automatically with molecular docking were in agreement with information obtained from existing X-ray structures. Both alignment methods produced statistically significant CoMFA and CoMSIA models, with the best q(2) value being 0.649 and the best predictive r(2) being 0.754. The manual alignment gave statistically higher values, whereas the automated alignment gave more robust models for predicting the activities of an external inhibitor set. Both models utilized similar amino acids in the HIV-1 PR active site, supporting the idea that hydrogen bonds form between an inhibitor and the backbone carbonyl oxygens of Gly48 and Gly48' and also the backbone NH group of Asp30, Gly48, Asp29', and Gly48' of the enzyme. These results suggest that an automated inhibitor alignment can yield predictive 3D QSAR models that are well comparable to manual methods. Thus, an automated alignment method in creating 3D QSAR models is encouragable when a well-characterized structure of the target protein is available.     

Improved 3D-QSAR prediction by multiple-conformational alignment: A case study on PTP1B inhibitors

Three-dimension quantitative structure activity relationship (3D-QSAR) was one of the major statistical techniques to investigate the correlation of biological activity with structural properties of candidate molecules, and the accuracy of statistic greatly depended on molecular alignment methodology. Exhaustive conformational search and successful conformational superposition could extremely improve the predictive accuracy of QSAR modeling. In this work, we proposed a solution to optimize QSAR prediction by multiple-conformational alignment methods, with a set of 40 flexible PTP1B inhibitors as case study. Three different molecular alignment methods were used for the development of 3D-QSAR models listed as following: (1) docking-based alignment (DBA); (2) pharmacophore-based alignment (PBA) and (3) co-crystallized conformer-based alignment (CCBA). Among these three alignments, it was indicated that the CCBA was the best and the fastest strategy in 3D-QSAR development, with the square correlation coefficient (r²) and cross-validated squared correlation coefficient (q²) of comparative molecular field analysis (CoMFA) were 0.992 and 0.694; the r² and q² of comparative molecular similarity indices analysis (CoMSIA) were 0.972 and 0.603, respectively. The alignment methodologies used here not only generated a robust QSAR model with useful molecular field contour maps for designing novel PTP1B inhibitors, but also provided a solution for constructing accurate 3D-QSAR model for various disease targets. Undoubtedly, such attempt in QSAR analysis would greatly help us to understand essential structural features of inhibitors required by its target, and so as to discover more promising chemical derivatives.     

Three-dimensional Quantitative Structure-Activity Relationship Analyses of a Series of Butenolide ETA Antagonists

Two kinds of Three-dimensional Quantitative Structure-activity Relationship(3D-QSAR) methods,comparative molecular filed analysis(CoMFA) and comparative molecular similarity indices analysis (CoMSIA) ,were applied to analyze the structure-activity relationship of a series of 63 butenolide ETA selective antagonists with respect to their inhibition against human ETA receptor,The CoMFA and CoMSIA models were developed for the conceivable alignment of the molecules based on a template structure from the crystallized data.The statistical results from the initial orientation of the aligned molecules show that the 3D-QSAR model from CoMFA(q^2=0.543) is obviously superior to that from the conventional CoMSIA(q^2=0.407).In order to refine the model,all-space search (ASS) was applied to minimize the field sampling process.By rotating and translating the molecular aggregate within the grid systematically,all the possible samplings of the molecular fields were tested and subsequently the one with the highest q^2 was picked out .The comparison of the sensitivity of CoMFA and CoMSIA to different space orientation shows that the CoMFA q^2 values are more sensitive to the translations and rotations of the aligned molecules with respect to the lattice than those of CoMSIA.The best CoMFA model from ASS was further refined by the region focused technique.The high quality of the best model is indicated by the high corss-validated correlation and the prediction on the external test set.The CoMFA coefficient contour plots identify several key features that explain the wide range of activities,which may help us to design new effective ETA selective antagonists.     

Three-dimensional quantitative structure-activity relationship (3D QSAR) and pharmacophore elucidation of tetrahydropyran derivatives as serotonin and norepinephrine transporter inhibitors

Three-dimensional quantitative structure-activity relationship (3D QSAR) using comparative molecular field analysis (CoMFA) was performed on a series of substituted tetrahydropyran (THP) derivatives possessing serotonin (SERT) and norepinephrine (NET) transporter inhibitory activities. The study aimed to rationalize the potency of these inhibitors for SERT and NET as well as the observed selectivity differences for NET over SERT. The dataset consisted of 29 molecules, of which 23 molecules were used as the training set for deriving CoMFA models for SERT and NET uptake inhibitory activities. Superimpositions were performed using atom-based fitting and 3-point pharmacophore-based alignment. Two charge calculation methods, Gasteiger-Hückel and semiempirical PM3, were tried. Both alignment methods were analyzed in terms of their predictive abilities and produced comparable results with high internal and external predictivities. The models obtained using the 3-point pharmacophore-based alignment outperformed the models with atom-based fitting in terms of relevant statistics and interpretability of the generated contour maps. Steric fields dominated electrostatic fields in terms of contribution. The selectivity analysis (NET over SERT), though yielded models with good internal predictivity, showed very poor external test set predictions. The analysis was repeated with 24 molecules after systematically excluding so-called outliers (5 out of 29) from the model derivation process. The resulting CoMFA model using the atom-based fitting exhibited good statistics and was able to explain most of the selectivity (NET over SERT)-discriminating factors. The presence of −OH substituent on the THP ring was found to be one of the most important factors governing the NET selectivity over SERT. Thus, a 4-point NET-selective pharmacophore, after introducing this newly found H-bond donor/acceptor feature in addition to the initial 3-point pharmacophore, was proposed.     

Insight into the structural requirements of narlaprevir-type inhibitors of NS3/NS4A protease based on HQSAR and molecular field analyses

In the life cycle of hepatitis C virus (HCV), NS3/NS4A protease has been proved to play a vital role in the replication of the HCV virus. Narlaprevir and its derivatives, the inhibitors of NS3/NS4A, would be potentially developed as important anti-HCV drugs in the future. In this study, quantitative structure-activity relationship (QSAR) analyses for 190 narlaprevir derivatives were conducted using comparative molecular field analysis (CoMFA), comparative molecular indices analysis (CoMSIA) and hologram quantitative structure-activity relationship (HQSAR) techniques. Both of the best CoMFA and HQSAR models showed statistical significance for the training set and good predictive accuracy for the test set, which strongly manifested the robustness of the CoMFA and HQSAR models. The CoMFA contour maps and the HQSAR contribution maps were both presented. Furthermore, based on the essential factors for ligand binding derived from the QSAR models, sixteen new derivatives were designed and some of them showed higher inhibitory activities confirmed by our models and molecular docking studies. General speaking, this study provides useful suggestions for the design of potential anti-HCV drugs.     

3D-QSAR studies on indole and 7-azoindole derivatives as ROCK-2 inhibitors: An integrative computational approach

Rho Kinases (ROCK) has been found to regulate a wide range of fundamental cell functions such as contraction, motility, proliferation, and apoptosis. Recent experiments have defined new functions of ROCKs in cells, including centrosome positioning and cell-size regulation, which might contribute to various physiological and pathological states. In this study, we have performed pharmacophore modeling and 3D QSAR studies on a series of 36 indoles and 7-azoindoles derivatives as ROCK2 inhibitors to elucidate the structural variations with their inhibitory activities. Ligand based CoMFA and CoMSIA models were generated based on three different alignment methods such as systematic search, simulated annealing and pharmacophore. A total of 15 CoMFA models and 27 CoMSIA were generated using different alignments. One model from each alignment is selected based on the statistical values. Contour maps of the selected models were compared, analysed and reported. The 3D QSAR study revealed that electro positive group linked to the methoxy-benzene ring position of the structure will enhance the biological activity and bulkier substitutions are preferred in the methyl dihydroindole region. Also, it is found that the hydrogen bond donor substituted at the R₁ position enhances the inhibitory activity. In future, this study would give proper guidelines to further enhance the activity of novel inhibitors for ROCK2.     

基于岭回归和SVM的高维特征选择与肽QSAR建模

岭回归估计权重绝对值在一定程度上体现了对应特征作用大小, 据此发展了基于岭回归(RR)和支持向量机(SVM)的高维特征选择算法. 对苦味二肽(BTT)和细胞毒性T淋巴细胞(CTL)表位9 肽两个肽体系, 以氨基酸的531 个物理化学性质参数直接表征肽结构, 各获得1062、4779 个初始特征; 对训练集, 初始特征以岭回归排序后序贯引入, 当SVM留一法交叉测试(LOOCV)的均方误差(MSE)显著上扬时终止, 最后以多轮末尾淘汰进一步精筛, 分别获得7、18个物理化学意义明确的保留特征. 基于保留特征与支持向量回归(SVR), 对训练集建立定量构效关系(QSAR)模型, 预测独立测试集, 其拟合精度、留一法交叉测试精度、独立预测精度均优于现有文献报道结果. 新方法运行速度快, 选取的特征物理化学意义明确, 解释性强, 在肽、蛋白质定量构效关系建模等高维数据回归预测领域有较广泛应用前景.