hERG钾离子通道和乳腺癌耐药蛋白抑制剂的理论预测研究

发布时间：2018-05-15 04:33

本文选题：ADMET + hERG钾离子通道　；参考：《浙江大学》2017年硕士论文

【摘要】：候选药物分子药代动力学性质(吸收、分布、代谢、排泄,ADME)和毒性(T)的优劣是药物研发是否成功的关键因素,因此在药物开发的早期对化合物的ADMET进行评价和优化具有重要的意义。本论文拟围绕hERG钾离子通道和乳腺癌耐药蛋白这两个与ADMET密切相关的蛋白展开研究,构建hERG钾离子通道和乳腺癌耐药蛋白的抑制剂预测模型,并研究受体-配体相互作用机制。阻碍hERG钾离子通道可能会导致长QT综合症、心律失常甚至猝死,因此为减少药物潜在的心脏毒性的风险而对hERG编码的钾离子通道毒性预测是十分必要的。在第一部分中,我们首先构建大量的药效团模型,然后使用递归分割方法确定了多个对hERG钾离子通道抑制剂和非抑制剂具有最优分类效果的药效团模型,最后通过机器学习方法(支持向量机、朴素贝叶斯)构建了基于多药效团的分类预测模型。最优的支持向量机模型对训练集的预测精度为84.7%,对测试集和外部验证集的预测精度均为82.1%；该模型能准确预测测试集中83.6%的抑制剂和78.2%的非抑制剂。此外,我们对重要的药效团模型进行了聚类分析,并通过分析具有代表性的药效团来描述hERG钾离子通道与配体间的多种相互作用机制。多药耐药现象是当今治疗恶性肿瘤失败的主要原因之一,乳腺癌耐药蛋白在多药耐药性起着至关重要的作用,其过量表达可能会导致药物达不到预期的疗效。在第二部分中,我们首先收集了860个乳腺癌耐药蛋白抑制剂和非抑制剂,并通过模拟退火和随机森林方法从大量的分子描述符中进行特征选择,并确定了36个重要的分子描述符；然后,基于最优描述符集和不同的分子指纹构建了朴素贝叶斯分类预测模型。结果表明：基于最优描述符集和分子指纹LCFP 4所构建的分类模型达到最好的预测效果,对训练集的预测精度为90.1%,对测试集的预测精度为94.2%,对外部验证集的预测精度为93.3%。此外,我们分析了对朴素贝叶斯分类模型贡献最大和最小的重要结构片段,深入探讨了BCRP抑制作用中起关键作用的结构片段。
[Abstract]:The advantages and disadvantages of molecular pharmacokinetic properties (absorption, distribution, metabolism, excretion, ADMEand toxicity) of candidate drugs are key factors for the success of drug development. Therefore, it is of great significance to evaluate and optimize the ADMET of compounds in the early stage of drug development. In this paper, we study the two proteins closely related to ADMET, hERG potassium channel and breast cancer resistance protein, and construct the inhibitor prediction model of hERG potassium channel and breast cancer resistance protein, and study the mechanism of receptor-ligand interaction. Blocking the potassium channel of hERG may lead to long QT syndrome arrhythmia and even sudden death so it is necessary to predict the potassium channel toxicity encoded by hERG in order to reduce the risk of potential cardiac toxicity of the drug. In the first part, we first construct a large number of pharmacophore models, and then we use recursive segmentation method to determine a number of pharmacophore models with optimal classification effects for hERG potassium channel inhibitors and non-inhibitors. Finally, a classification and prediction model based on multi-pharmacophore is constructed by means of machine learning (support vector machine, naive Bayes). The prediction accuracy of the optimal support vector machine model for the training set is 84.7, the prediction accuracy for the test set and the external verification set is both 82.1, and the model can accurately predict 83.6% of the inhibitors and 78.2% of the non-inhibitors in the test set. In addition, the important pharmacophore models were clustered and the representative pharmacophore groups were analyzed to describe the interaction mechanisms between hERG potassium channels and ligands. Multidrug resistance (MDR) is one of the main causes of failure in the treatment of malignant tumors. Multidrug resistance protein plays an important role in MDR, and its overexpression may lead to drug failure. In the second part, we first collected 860 breast cancer resistant protein inhibitors and non-inhibitors, and used simulated annealing and random forest methods to select features from a large number of molecular descriptors. 36 important molecular descriptors are determined, and then a naive Bayesian classification and prediction model is constructed based on the optimal descriptor set and different molecular fingerprints. The results show that the classification model based on the optimal descriptor set and the molecular fingerprint LCFP _ 4 has the best prediction effect. The prediction accuracy for the training set is 90.1, for the test set is 94.2 and for the external verification set is 93.3. In addition, we analyze the important structural fragments that contribute the most and least to the naive Bayesian classification model, and discuss in depth the structural fragments that play a key role in the inhibition of BCRP.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R96

【参考文献】