机器学习方法用于选择性环氧化酶-2抑制剂活性预测模型的建立

与传统的非甾体类消炎药相比,选择性环氧化酶-2抑制剂具有无胃肠道粘膜损伤,溃疡和肾功能障碍等严重的副作用,设计选择性环氧化酶-2抑制剂具有重要意义。本文用支持矢量学习机和神经网络两种机器学习方法建立选择性环氧化酶-2抑制剂的活性预测模型,以期为选择性环氧化酶-2抑制剂药物的合成提供先导化合物。我们将467个环氧化酶-2抑制剂用Kennard-Stone方法分为训练集,验证集和独立测试集,对每一抑制剂分子我们计算了463个包含组成描述符和拓扑描述符的分子描述符来表征其分子结构,并通过F-Score方法选取最重要的分子描述符用于分类模型的建立。结果表明,SVM方法通过变量筛选后具有很好的预测能力,其预测正确率达到93.30%。     

Homology model-guided 3D-QSAR studies of HIV-1 integrase inhibitors

In the present study, we report the exploration of binding modes of potent HIV-1 integrase (IN) inhibitors MK-0518 (raltegravir) and GS-9137 (elvitegravir) as well as chalcone and related amide IN inhibitors we recently synthesized and the development of 3D-QSAR models for integrase inhibition. Homology models of DNA-bound HIV-1 IN were constructed on the basis of the X-ray crystal structure of the foamy virus IN-DNA complex (PDB ID: 3L2T ) and used for docking. The binding modes of raltegravir and elvitegravir in our homology models are in accordance with those in the foamy virus structure revealing interactions important for inhibitor-IN binding. To gain further insights into the structural requirements for IN inhibition, three-dimensional quantitative structure activity relationship (3D-QSAR) studies were conducted using raltegravir, elvitegravir, and their analogs; our synthesized 3-keto salicylic acid IN inhibitor series; as well as other structurally related HIV-1 IN inhibitors. In the first part of the study with 103 compounds, atom-fit alignments, I and II, and docking-based alignment, III, were used to develop 3D-QSAR models 1, 2, and 3, respectively, each comprising comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) 3D-QSARs. This initial analysis indicated that the docking-based (structure-based) model 3 performed better than the atom-fit (ligand-based) models 1 and 2, in terms of statistical significance and robustness. Thus, the docking-based alignment was then subsequently used with an expanded data set of 296 compounds for building a more comprehensive 3D-QSAR, model 4. Model 4 afforded good q2 values of 0.70 and 0.75 for CoMFA and CoMSIA 3D-QSARs, respectively, and showed good predictive performance on an external validation test set of 59 compounds with predictive r2 values up to 0.71. The HIV IN-DNA homology model of biological relevance and the comprehensive 3D-QSAR models developed in the present study provide insights and new predictive tools for structure-based design and optimization of IN inhibitors.     

基于分子对接的6-萘甲基取代HEPT类HIV-1逆转录酶抑制剂构效关系研究

采用分子对接方法得到了一系列6-萘甲基取代HEPT类逆转录酶抑制剂分子与HIV-1逆转录酶复合物模型,从中抽取出抑制剂分子的活性构象,进一步应用CoMFA和CoMSIA方法建立了具有较好预测能力的3D-QSAR模型,深入探讨了这些化合物的定量构效关系,为进一步的药物设计奠定了良好的基础.另外,以化合物13及其相应的β异构体24为代表,结合量子化学从头算分子轨道理论方法考察了它们的前线轨道,为阐明α和β系列化合物的活性差异提供了理论依据.     

Three-dimensional Quantitative Structure-activity Relationship Models of HIV-1 Integrase Inhibitors of DKAs

As one of the three viral encoded enzymes of HIV-1 infection, HIV-1 integrase has become an attractive drug target for the treatment. Diketoacid compounds (DKAs) are one kind of potent and selective inhibitors of HIV-1 IN. In the present work, two three-dimensional QSAR techniques (CoMFA and CoMSIA) were employed to correlate the molecular structure with the activity of inhibiting the strand transfer for 147 DKAs. The all-oritation search (AOS) and all-placement search (APS) were used to optimize the CoMFA model. The diketo and keto-enol tautomers of DKAs were also used to establish the CoMFA models. The results indicated that the enol was the dominant conformation in the HIV-1 IN and DKAs complexes. It can provide a new method and reference to identify the bioactive conformation of drugs by using QSAR analysis. The best CoMSIA model, with five fields combined, implied that the hydrophobic field is very important as well as the steric and electrostatic fields. All models indicated favorable internal validation. A comparative analysis with the three models demonstrated that the CoMFA model seems to be more predictive. The contour maps could afford steric, electrostatic, hydrophobic and H-bond information about the interaction of ligand-receptor complex visually. The models would give some useful guidelines for designing novel and potent HIV-1 integrase inhibitors.