老藥新用及細胞毒性分析與選擇性偏優化

Cytotoxicity Data Analysis and Biased Optimization for Selectivity in Drug Development and Drug Repurposing

張嘉芸(Chia-Yun Chang)

指導教授 ： 曾宇鳳 ； 孔繁璐

國立臺灣大學/醫學院/藥學研究所/博士(2013年)

https://doi.org/10.6342/NTU.2013.10206

摘要

本論文中提出三項研究主題，第一章說明，N-甲基-D-天冬氨酸(NMDA)受體已被發現參與精神分裂症相關的神經病理，並已知D-氨基酸氧化酶(DAAO)在NMDA受體的激活過程中發揮重要的作用，因此DAAO被推斷與精神分裂的致病機轉有關。在本研究中，我們利用老藥新用的方式，針對已知的藥物進行DAAO可能的抑制劑篩選，透過以結構為基礎(structure-based)的虛擬篩選(virtual screening)最後進行了八個候選化合物的DAAO酵素實驗，其中有四個化合物對DAAO的抑制能力其IC50在1~10 μM之間，又以5-O-去甲基-奧美拉唑(5-O-desmethyl-omeprazole)有最佳的抑制力，其為一種用於治療消化不良、潰瘍和胃食道逆流藥物的代謝物，他的IC50值為1.19 μM。
第二章中，我們針對化合物的選擇性和藥物的修改方向透過基於配體(ligand-based)的定量構效關係(QSAR)模型組合，發展一個偏優化(biased optimization)的方法。多目標藥物具有治療多種疾病的潛力，但可能也具有不良的副作用，因此，將化合物的作用導向我們想要的方向變的十分重要，我們選擇了兩系列作用於HER2/EGFR和CDKs/CK1的化合物，說明QSAR不只可以針對藥物的專一性進行優化，也可應用於藥物的選擇性，QSAR回歸模型的係數指出分子上哪些位置和特性對於分子的選擇性有決定性的影響。我們還針對分子結合到相關受體上的情況進行分析，說明基於配體的方法仍可以反映出和受體結合後的差異，並可避免相關結晶結構未知的問題。
在第三章使用支持向量機(SVM)、各種分子特徵及不同取樣方法的組合對於多個化合物針對Jurkat這個細胞株進行細胞毒性的分類與預測，該細胞毒性為不平衡的數據組成，也就是少數的化合物具有細胞毒性，多數則無，這將不利預測模型的建構。我們透過SVM和各種分子特徵，包括：4D-FPs、 MOE (1D, 2D and 2½D)、noNP+MOE和CATS2D等方法，和前人使用隨機森林(random forest)法的CATS2D分子特徵所建構出來的模型之預測利進行比較，使用SVM搭配超取樣(oversample)的方式，不論是對於數據的訓練或是額外的預測都有較佳的結果。

關鍵字

D-氨基酸氧化酶虛擬篩選 ； 老藥新用 ； 偏優化 ； 四維定量構效關係 ； 選擇性 ； 不平衡數據組成 ； 細胞毒性模型 ； 分類預測模型

並列摘要

Three works are proposed in this dissertation. In the first chapter, the N-methyl-D-aspartic acid (NMDA) receptor has previously been reported to be involved in the neuropathology of schizophrenia. D-amino acid oxidase (DAAO) has been found to play an important role in the activation process of the NMDA receptor and consequently has also been hypothesized to be involved in the pathogenesis of schizophrenia. To determine if existing drugs and compounds could be utilized as potential DAAO inhibitors through drug repurposing, a structure-based virtual screening was performed utilizing the DrugBank database. Eight commercially available compounds were identified and selected for further analysis with a DAAO enzymatic assay, with four of these compounds confirmed to have DAAO inhibitory activity with IC50 values ranging from 1 to 10 μM. 5-O-desmethyl-omeprazole, a metabolite of omeprazole, an agent widely used to treat dyspepsia, peptic ulcer disease and gastroesophageal reflux disease, was identified in this study as the most potent inhibitor of DAAO with an IC50 value of 1.19 μM.
In Chapter 2, the purpose of this study was to develop a biased optimization method that responds to the selectivities of compounds and to modify the drug optimization process through simple ligand-based quantitative structure–activity relationship (QSAR) model combinations. Multi-target drugs not only have the potential to treat multiple diseases but also possess a higher probability of undesirable side effects. Therefore, we sought to develop inhibitors that possess only the activities we deemed important. We chose two series of compounds to act on HER2/EGFR and CDKs/CK1 and demonstrated that the QSAR methodology for drug specificity can be extended to drug selectivity. The coefficients in the regression models indicate which molecular positions and replacements are important for determining selectivity. We also analyzed ligand-binding conditions to demonstrate that the ligand-based method describes differences in the receptor structures and avoids estimation problems that arise when the crystal structures of the complexes are unknown.
In Chapter 3, the research presented in this study focuses on the use of support vector machines, a machine learning method, various classes of molecular descriptors and different sampling techniques to overcome overfitting to classify compounds for cytotoxicity with respect to the Jurkat cell line. The cell cytotoxicity data set is imbalanced (few active compounds and very many inactive compounds) and the ability of the predictive modeling methods is adversely affected in these situations. The SVM models were constructed from 4D-FPs, MOE (1D, 2D and 2½D), noNP+MOE and CATS2D trial descriptors pools and compared to the predictive abilities of CATS2D-based random forest models. Compared to previous studies, the SVM models built from oversampled data sets exhibited better predictive abilities for the training and external test sets.

並列關鍵字

DAAO ； Virtual screening ； Drug repurposing ； Biased optimization ； 4D-QSAR, Selectivity ； Imbalanced data ； Cytotoxicity modeling ； Classification predictive model

參考文獻

1. Field, J. R.; Walker, A. G.; Conn, P. J., Targeting glutamate synapses in schizophrenia. Trends in molecular medicine 2011, 17 (12), 689-98.

2. Paoletti, P.; Neyton, J., NMDA receptor subunits: function and pharmacology. Current opinion in pharmacology 2007, 7 (1), 39-47.

3. Kleckner, N. W.; Dingledine, R., Requirement for glycine in activation of NMDA-receptors expressed in Xenopus oocytes. Science 1988, 241 (4867), 835-7.

4. Coyle, J. T.; Tsai, G., The NMDA receptor glycine modulatory site: a therapeutic target for improving cognition and reducing negative symptoms in schizophrenia. Psychopharmacology 2004, 174 (1), 32-8.

5. Schell, M. J., The N-methyl D-aspartate receptor glycine site and D-serine metabolism: an evolutionary perspective. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences 2004, 359 (1446), 943-64.