【摘要】：極光激酶A是一種廣為認(rèn)可的抗癌藥物設(shè)計(jì)的靶標(biāo),其抑制劑的設(shè)計(jì)研發(fā)有重要的研究意義。利用計(jì)算機(jī)輔助藥物設(shè)計(jì)的各種方法,通過對(duì)已知的極光激酶A抑制劑進(jìn)行構(gòu)效關(guān)系研究,包括建立分類模型和定量預(yù)測(cè)模型,以及基于小分子配體化合物的虛擬篩選,可以從小分子化合物數(shù)據(jù)庫中篩選出具有抑制極光激酶A的苗頭化合物。本課題研究主要包含為以下三方面的研究?jī)?nèi)容： (1)利用自組織神經(jīng)網(wǎng)絡(luò)和支持向量機(jī)法建立了極光激酶A抑制劑生物活性的分類模型。對(duì)已知結(jié)構(gòu)和活性的1463個(gè)極光激酶A抑制劑,基于生物活性為閾值,選擇了29個(gè)ADRIANA.Code結(jié)構(gòu)特征符,用自組織神經(jīng)網(wǎng)絡(luò)和支持向量機(jī)法分別建立分類模型ModelA1和ModelA2。其中ModelA1對(duì)訓(xùn)練集和測(cè)試集的預(yù)測(cè)正確率分別為91.19%和86.10%,ModelA2對(duì)訓(xùn)練集和測(cè)試集的預(yù)測(cè)正確率分別為94.13%和86.11%。此外還利用ECFP4指紋圖譜分析抑制劑生物活性與其子結(jié)構(gòu)的關(guān)系。 (2)利用多元線性回歸和支持向量機(jī)建立了極光激酶A抑制劑生物活性的定量預(yù)測(cè)模型研究。根據(jù)抑制劑的三種不同酶學(xué)活性測(cè)定方式,將抑制劑分成三個(gè)子集,分別包括356/302/279個(gè)化合物。每個(gè)子集用隨機(jī)和自組織神經(jīng)網(wǎng)絡(luò)兩種方法劃分訓(xùn)練集和測(cè)試集,再分別用多元線性回歸法和支持向量機(jī)建立模型。并對(duì)每個(gè)子集建立了四個(gè)定量預(yù)測(cè)模型。所有模型對(duì)測(cè)試集的活性值定量預(yù)測(cè)相關(guān)系數(shù)R均不小于0.77。模型均通過隨機(jī)分布法檢驗(yàn),顯示出良好的預(yù)測(cè)能力。 (3)基于極光激酶A抑制劑為配體的虛擬篩選研究。依據(jù)分子三維形狀相似性和靜電相似性,主要利用ROCs和EON軟件,探尋建立不同目標(biāo)查詢式(query)的方法,最終選定化合物vx680為query,進(jìn)行虛擬篩選,從近五百萬個(gè)小分子數(shù)據(jù)庫中得到500個(gè)與目標(biāo)查詢式最相似的化合物。這些化合物通過最優(yōu)分類模型和最優(yōu)定量預(yù)測(cè)模型逐一篩選后,得到預(yù)測(cè)活性值小于10nM的23個(gè)化合物。其被認(rèn)為是具有抑制極光激酶A的苗頭化合物。 本研究通過對(duì)極光激酶A抑制劑生物活性的分類和定量預(yù)測(cè)研究,建立了一系列的定性和定量預(yù)測(cè)模型。通過對(duì)極光激酶A抑制劑的三維分子形狀和靜電相似的虛擬篩選研究,找到了一些潛在的抑制極光激酶A的苗頭化合物。
[Abstract]:Aurora kinase A is a widely recognized target for the design of anticancer drugs, and the design and development of its inhibitors is of great significance. By using various methods of computer-aided drug design, the structure-activity relationships of known Aurora inhibitors were studied, including the establishment of classification models and quantitative prediction models, and virtual screening based on small molecular ligands. Seedling compounds with inhibition of aurora kinase A can be screened from a small molecular compound database. The main contents of this paper are as follows: (1) the classification model of the biological activity of Aurora kinase A inhibitors is established by using self-organizing neural network and support vector machine. For 1463 Aurora kinase A inhibitors with known structure and activity, based on the threshold of biological activity, 29 ADRIANA.Code structural characteristics were selected. The classification model ModelA1 and ModelA2. were established by using self-organizing neural network and support vector machine method, respectively. The prediction accuracy rate of ModelA1 for training set and test set is 91.19% and 86.10% respectively for training set and test set are 94.13% and 86.11% respectively. In addition, the relationship between the bioactivity of inhibitors and their substructures was analyzed by ECFP4 fingerprinting. (2) A quantitative prediction model for the bioactivity of Aurora kinase A inhibitors was established by using multiple linear regression and support vector machine. The inhibitors were divided into three subsets according to three different enzymatic activities of inhibitors, consisting of 356 / 302 / 279 compounds. Each subset is divided into training set and test set by stochastic and self-organizing neural network methods, and then the model is established by multivariate linear regression method and support vector machine, respectively. Four quantitative prediction models are established for each subset. The correlation coefficient R of all models for quantitative prediction of activity values of test sets is not less than 0.77. The models were tested by random distribution method and showed good predictive ability. (3) Virtual screening study based on Aurora kinase A inhibitor as ligand. According to the similarity of three-dimensional and electrostatic shapes of molecules, the methods of establishing different target query (query) were explored by using ROCs and EON software. Finally, the compound vx680 was selected for virtual screening of query,. 500 compounds most similar to the target query were obtained from nearly five million small molecular databases. These compounds were screened one by the optimal classification model and the optimal quantitative prediction model, and 23 compounds whose predictive activity values were less than 10nM were obtained. It is thought to be a seeding compound that inhibits auroral kinase A. In this study, a series of qualitative and quantitative prediction models were established based on the classification and quantitative prediction of the biological activity of Aurora kinase A inhibitors. Based on the virtual screening of the three-dimensional molecular shape and electrostatic similarity of Aurora kinase A inhibitors, some potential antagonist compounds of aurora kinase A were found.
【學(xué)位授予單位】：北京化工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2014
【分類號(hào)】：R91-39

【參考文獻(xiàn)】

相關(guān)期刊論文 前10條

1 丁世飛;齊丙娟;譚紅艷;;支持向量機(jī)理論與算法研究綜述[J];電子科技大學(xué)學(xué)報(bào);2011年01期

2 黃琦;康宏;張端峰;盛振;劉琦;朱瑞新;曹志偉;;基于配體、受體和復(fù)合物指紋的虛擬篩選方法比較[J];化學(xué)學(xué)報(bào);2011年05期

3 楊銳;韓葳葳;王嵩;;極光激酶A抗癌抑制劑的高通量篩選及對(duì)接研究[J];化學(xué)學(xué)報(bào);2011年12期

4 任偉;孔德信;;基于配體和受體的藥物設(shè)計(jì)方法的對(duì)應(yīng)性[J];生命科學(xué)儀器;2009年03期

5 安麗英;相玉紅;張卓勇;胡文祥;;定量構(gòu)效關(guān)系研究進(jìn)展及其應(yīng)用[J];首都師范大學(xué)學(xué)報(bào)(自然科學(xué)版);2006年03期

6 唐，

本文編號(hào)：2230132