【作者】 夏昆华；

【导师】 周鲁；

【作者基本信息】 四川大学 ， 工业催化， 2004， 硕士

【摘要】 喹诺酮类抗菌素是一类全人工合成的抗菌药物,自从1962年萘啶酸问世以来,已有数以千计的喹诺酮类化合物得以合成,至今,上市的喹诺酮类药物已达几十种。宽广的抗菌谱、良好的药动学性质、高效低毒等特点使喹诺酮类抗菌素成为了临床上使用最为广泛的抗菌药物之一。但是,喹诺酮类化合物也有很多不良的临床反应和毒副作用,急需设计和寻找出抗菌活性更强、安全性更高的喹诺酮类新药。对喹诺酮类化合物的药效学/药动学性质、抗菌作用机理和化合物分子结构之间的关系做深入的研究,可以为设计和寻找新药提供指导性的参考和帮助。本文利用量子化学和神经网络方法,分别进行了喹诺酮类化合物的药效学、药动学参数与其分子结构参数之间的相关性分析。我们选用了已知对金葡萄球菌和大肠杆菌的最小抑菌浓度的110个喹诺酮类化合物来研究其药效学性质。我们分别计算了110个化合物的18个结构参数:分子总能量、分子键能、分子离解能、电子能、核-核间排斥能 、生成热、单点偶极矩、偶极矩、7-位取代基的净电荷、分子表面积、分子体积、水合能、疏水性参数、分子折射率、分子极化率、分子摩尔质量、分子最高占有轨道能量和分子最低空轨道能量。首先,对这18个结构参数之间进行了自相关性分析,把相关系数大于0.75的参数归为一组,共分为了9组,然后通过试算的方法最终选择了6个结构参数作为网络的输入,分别对金葡萄球菌和大肠杆菌的最小抑菌浓度进行神经网络建模。6个结构参数分别是:生成热、偶极矩、7-位取代基的净电荷、分子表面积、水合能、分子最高占有轨道能量。网络参数选择如下:最大迭代次数为5000,第一隐含层节点数分别为20和18,第二隐含层节点数分别为32和28,目标误差为0.02、学习速率的初值选定为0.01、学习速率增加比例为1.05、学习速率减少比例为0.8、动量常数为0.9。我们随机选取了100个样本作为训练集,其余10个样本作为预测集来检验网络的性能。计算结果显示,对金葡萄<WP=3>球菌的最小抑菌浓度学习正确率为64%、预测正确率为70%;对大肠杆菌的最小抑菌浓度学习正确率为74%、预测正确率为60%。另外一方面,我们选择了21个已知药动学参数:达峰浓度、血药浓度-时间曲线下面积和半衰期的喹诺酮类化合物进行药动学参数和分子结构参数的相关性分析。同样,我们分别计算了化合物的上述18个结构参数,并用同样的方法筛选出了6个结构参数作为网络的输入:偶极矩、7-位取代基的净电荷、分子体积、水合能、疏水性参数、分子最高占有轨道能量。通过试算,我们确定了网络的其他参数为:最大迭代次数为5000,第一隐含层节点数分别为16,第二隐含层节点数分别为26,目标误差为0.01、学习速率的初值选定为0.01、学习速率增加比例为1.05、学习速率减少比例为0.8、动量常数为0.9。我们随机取20个化合物作为训练集,并对训练集中的化合物进行Leave-one-out分析,对剩下的一个化合物的三个药动学参数进行预测。结果表明,网络对三个药动学参数的学习正确率分别为:85%、75%和95%,Leave-one-out预测正确率分别为:70%、60%和80%,对三个药动学参数预测的相对误差为:18.44%、14.46%和-8.25%。结果表明我们对喹诺酮类化合物药效学/药动学参数的网络建模是合理的,我们可以用改网络模型来准确预测喹诺酮类化合物的药效学/药动学参数。在计算结果的基础上,我们从抗菌机理的角度分析了影响喹诺酮类化合物药效学/药动学性质的6个结构参数与分子结构之间的关系,探讨了不同结构参数对抗菌作用机理以及体内吸收、分布方面的影响。从而为分析喹诺酮类药物的药效学/药动学性质以及新药设计提供了思路和参考意见。更多还原

【Abstract】 Quinolones is a kind of antibacterial synthesized artificially. There are thousands of compounds of quinolones have been synthesized since 1962, and now there are dozens drugs of quinolones appear on maket. Antibacterial quinolones had a wide clinical application because of its high activity, good pharmacokinetics quality and low toxicity. But quinolones can redounds many toxicity and side effect, so it is necessary to devise and search better drugs of quinolones, which have higher activity, better PK quality and lower toxicity. Study the relationship of pharmacodynamics and pharmacokinetics quality of quinolones, action mechanisms and the molecule structure plays crucial roles in the design of more potent drugs. In this paper, a correlative analysis is given between the pharmacodynamics/pharmacokinetics parameters and the molecule structure parameters by use of quantum chemistry and neural network. 110 compounds are selected and their minimum inhibition concentrations to Staphylococcus aureus and Escherichia coli have been determined. The 18 molecule structure parameters have been calculated: total energy, binding energy, isolated energy, electronic energy, core-core energy, heat of formulation, point dipole moment, dipole moment, 7-net charge, surface area, volume of molecule, hydration energy, parameter of distant water, refractive index of molecule, polarizational rate, mole weight of molecule, EHOMO and ELUMO. We firstly analyzed the relativity of the 18 parameters, and distribute the relativity coefficient larger than 0.75 to a same group. There are 9 groups at all. Then, we use the method of try, choose the 6 parameters for the input of net to build networks separately. The 6 parameters is heat of formulation, dipole moment, 7-net charge, surface area, hydration energy and EHOMO. The network’s structural parameters are described as: the maximal cycles are 5000, the nodes of the first-hidden layer are 20 and 18, the ones of second-hidden layer are 32 and 28, target error is 0.02, original learning rate is 0.01 and its increasing ratio is 1.05 and its decreasing ratio is 0.8, the momentum <WP=5>factor is 0.9. By this network, 100 compounds are selected stochastically as the training gather and the residual 10 ones as the prediction gather for validation our network. The correct ratio of being learned and predicted of the minimum inhibition concentrations to Staphylococcus aureus is 64% and 70%; the correct ratio of being learned and predicted of the minimum inhibition concentrations to Escherichia coli is 74% and 60%. On the other hand, 21 compounds are selected to study the relationship between the molecule structure parameters of quinolones and PK parameters: Cmax, AUC, T1/2. Equally, we calculate the 18 molecule structure parameters of the 21 compounds and choose 6 parameters for input of the net: dipole moment, 7-net charge, volume of molecule, hydration energy, parameter of distant water and EHOMO. The other parameters of network is: the maximal cycles are 5000, the nodes of the first-hidden layer are 16, the ones of second-hidden layer are 26, target error is 0.01, original learning rate is 0.01 and its increasing ratio is 1.05 and its decreasing ratio is 0.8, the momentum factor is 0.9. 20 compounds are selected stochastically as the training gather and tested by leave-one-out method. The 3 pharmacokinetics parameters of the leave one have been predicted. The result shows, The correct ratio of being learned of the 3 PK parameters is 85%, 75% and 95%; The correct ratio of being predicted by leave-one-out of the3 pharmacokinetics parameters is 70%, 60% and 80%; and the relatively error of being predicted of the 3 pharmacokinetics parameters is 18.44%, 14.46% and –8.25%. All the results show the model between the pharmacodynamics/pharmacokinetics parameters and the structure parameters of quinolones molecules is reasonable, and we can use the networks to predicte the pharmacodynamics/pharmacokinetics parameters of antibacterial quinolones. Based on the results, we have analyzed the relationship of 6 stru更多还原