P-gp抑制剂和底物预测与分子动力学模拟

发布时间：2018-04-24 00:41

本文选题：P-gp + 定量构效关系　；参考：《重庆大学》2014年硕士论文

【摘要】：由P-糖蛋白(P-glycoprotein, P-gp)过度表达导致的肿瘤细胞的多药耐药性(multidrug resistance, MDR)，是肿瘤化疗失败的主要原因。因此，P-gp抑制剂预测和筛选已经成为肿瘤药物研发的一个重要内容。论文以P-gp为研究对象，采用定量构效关系、同源模建、分子对接和分子动力学模拟等相关理论方法，对P-gp抑制剂和底物进行了分类预测以及作用机理研究。论文研究结果对P-gp抑制剂相关药物研发具有重要的理论意义和应用价值。主要研究内容和研究结果： ①以小鼠P-gp晶体结构为模板构建了两个P-gp同源模型，并采用Amber力场对同源模型进行了能量优化。模型质量评价结果表明，，本文构建的P-gp同源模型在结构和能量上合理、可靠。 ②采用支持向量机(Support Vector Machine, SVM)和分子对接两种方法进行了P-gp抑制剂的预测研究。根据10-折交互验证结果，得到了仅包含3个描述子的最优分类模型，其对857个训练集样本的预测准确度、敏感度、特异度和马修斯相关系数分别为0.840、0.873、0.813和0.683；对418个测试集样本的预测准确度、敏感度、特异度和马修斯相关系数分别为0.868、0.938、0.738和0.704。Surflex-dock分子对接结果表明：对接得分可较好评价P-gp抑制剂/非抑制剂的相互作用，并具有较好的预测能力。进一步研究显示，最优SVM模型和分子对接两种预测方法对77.3%的样本具有一致性的预测结果，其预测准确度达到了0.931。与此同时，对接结果也显示Phe699、Phe303、Phe695、Phe945和Phe39是P-gp抑制剂的重要结合位点。 ③采用SVM和分子对接方法构建P-gp底物的分类模型。根据10-折交互验证结果，得到包含5个描述子的最优分类模型，其对1731个训练集样本的预测准确度、敏感度、特异度和马修斯相关系数分别为0.69、0.85、0.47和0.34，对193个测试集样本的预测准确度、敏感度、特异度和马修斯相关系数分别为0.71、0.84、0.49和0.35。分子对接结果表明：Surflex-dock对接得分不能有效地区分P-gp底物和非底物。进一步研究显示：P-gp底物和非底物的结合位点无显著性差异。 ④在P-gp同源模型基础上，进一步构建了P-gp的跨膜模型，并在此基础上进行了10ns的分子动力学模拟，以考察细胞膜和配体对P-gp构象的影响。研究结果显示：经10ns的分子动力学模拟，P-gp的跨膜模型在能量和结构上达到平衡状态。基于动力学的最低能量构象，论文重新对1929个P-gp底物/非底物进行了对接研究。研究结果显示：Surflex-dock对接得分仍无法有效区分P-gp底物和非底物。究其原因可能是P-gp底物结构和作用模式多样性造成的。
[Abstract]:The multidrug resistance (MDR) caused by the overexpression of P-gp is the main reason for the failure of chemotherapy. Therefore, the prediction and screening of P-gp inhibitors have become an important content in the development of tumor drugs. In this paper, using quantitative structure-activity relationship, homology modeling, molecular docking and molecular dynamics simulation, the P-gp inhibitors and substrates were classified and predicted and the mechanism of their action was studied. The results of this paper have important theoretical significance and application value for the research and development of P-gp inhibitor-related drugs. The main research contents and results are as follows: 1 two P-gp homologous models were constructed using mouse P-gp crystal structure as template, and the energy of the homologous models was optimized by Amber force field. The model quality evaluation results show that the P-gp homologous model constructed in this paper is reasonable and reliable in structure and energy. 2 the prediction of P-gp inhibitors was studied by support vector machine support Vector machine and molecular docking. Based on the results of 10-fold interactive verification, an optimal classification model containing only three descriptors is obtained. The prediction accuracy and sensitivity of 857 training set samples are obtained. The correlation coefficients of specificity and Matthews were 0.840 ~ 0.873n 0.813 and 0.683, respectively. The predictive accuracy and sensitivity of 418 test samples were analyzed. The correlation coefficients of specificity and Matthews were 0.868 / 0.938 and 0.738, respectively. The results of docking showed that the docking score could evaluate the interaction between P-gp inhibitors and non-inhibitors, and had a good predictive ability. Further studies show that the optimal SVM model and molecular docking methods have consistent prediction results for 77.3% samples, and the prediction accuracy is 0.931. At the same time, the docking results also showed that Phe699, Phe303, Phe695Phe945 and Phe39 were important binding sites for P-gp inhibitors. (3) the classification model of P-gp substrate was constructed by SVM and molecular docking. According to the results of 10-fold interactive verification, an optimal classification model containing five descriptors is obtained. The prediction accuracy and sensitivity for 1731 training set samples are obtained. The predictive accuracy, sensitivity, specificity and Matthews correlation coefficient of 193 test samples were 0.71 ~ 0.84 ~ 0.49 and 0.35, respectively. The correlation coefficients of specificity and Matthews were 0. 69 and 0. 34, respectively. The predictive accuracy, sensitivity, specificity and Matthews correlation coefficient of 193 test samples were 0. 71 ~ 0. 84 and 0. 35, respectively. The results of molecular docking show that the docking score of 1: Surflex-dock can not effectively distinguish P-gp substrate from non-substrate. Further studies showed that there was no significant difference in the binding sites between substrates and non-substrates. (4) based on the P-gp homology model, the transmembrane model of P-gp was further constructed, and the molecular dynamics simulation of 10ns was carried out to investigate the effect of cell membrane and ligand on the conformation of P-gp. The results show that the transmembrane model of P-gp is balanced in energy and structure by 10ns molecular dynamics simulation. Based on the lowest energy conformation of dynamics, the docking of 1929 P-gp substrates / non substrates was studied again. The results show that the P-gp substrates and the non-substrates can not be effectively distinguished by a: Surflex-dock docking score. The reason may be the diversity of P-gp substrate structure and mode of action.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：R91

【共引文献】