基于机器学习的天然产物抗肿瘤和免疫调节活性研究

发布时间：2018-08-08 16:43

【摘要】：天然产物在癌症化学疗法,免疫调节,神经保护,调节能量代谢,抗微生物以及心血管疾病治疗等方面起到了十分重要的作用。他们很可能会给我们提供很多先导化合物的结构,可以作为药物开发的模板用于多种疾病的预防与治疗。本文主要是对天然产物的抗肿瘤活性以及天然产物中的三萜皂苷的免疫调节活性进行研究。通过机器学习方法分别建立预测模型,并且找出能够起到描述天然产物相应生物活性的关键化学结构。本文的方法以及结果在以天然产物为基础的药物开发利用,疾病的治疗,以及天然产物的构效关系的研究中起到积极的促进的作用。(1)天然产物抗肿瘤的研究传统的天然产物构效关系的研究方法通常可以获得一些关键的结构特性,比如某些化学键或者化学基团。在天然产物抗肿瘤活性的研究中,通常首先确定这个结构特征对某个单一的癌症类型或者是细胞系的影响。在本文中,我们首次开发了一种机器学习方法去全面地预测天然产物对一系列癌症细胞系的抗肿瘤效果。这里我们同时用到了癌细胞系的基因组信息(基因表达)以及天然产物的化学结构特征(化学描述符)。当这些机器学习模型分别应用到两个数据集,训练集和独立的测试集时,这些预测模型展现出了非常好的预测准确性。此外,我们也证明了该方法对于两种天然产物,即姜黄素和白藜芦醇抗肿瘤活性的预测能力。这表明该方法能够有效地预测出这两种天然产物与多种癌症细胞系敏感性的一一对应的关系。同时把病人的基因组特征信息与天然产物的敏感性联系起来。基于以上结果,有理由认为该方法能够在癌症的预防治疗以及精确性医疗的发展中,对鉴定潜在的候选天然产物起到促进的作用。近年来,很多天然产物在实验室被分离纯化,并且被证明具有一定的抗肿瘤活性,例如喜树碱,长春花碱,信筒子醌,紫杉醇等等。这些天然来源的产物已在黑色素瘤,白血病,乳腺癌等多种癌症的治疗当中起到关键的作用。另外,越来越多基于天然产物结构的新的衍生物成为了抗肿瘤的候选药物。然而,用实验的方法来筛天然产物先导化合物是极为费时费力的。因此,应用计算的方法对其构效关系的研究是一个好的选择。传统的研究构效关系的方法大都是拥有化学背景的科研人员以先导化合物的基本骨架为基础,合成一系列的类似物,再根据经验来对它们的抗肿瘤活性进行预测。这些构效关系的研究方法只用结构信息,尝试对单个癌症细胞系或者单一组织类型的抑制进行预测。尽管已经取得了许多成果,该问题还远远没有被很好地解决。为了获得全面的分析,我们打算使用一种新的,同时基于化学结构和基因组信息的研究思路。这样可以一次同时检测多个己知的癌症细胞系对数百种天然产物的敏感性。我们首次使用癌细胞系的基因表达数据和天然产物结构的化学描述符建立了机器学习方法来预测癌细胞系对天然产物的反应。已有的训练集和测试集的实验结果证明我们的方法能够在预测数百个癌细胞系的敏感性时,获得比较满意的表现。也同时证明了结构特征与基因表达特征的结合,对天然产物的抗肿瘤活性具有重要的决定作用,该研究项目对药物反应的预测以及发现新的天然来源的候选药物有一定的指导和帮助。(2)三萜皂苷免疫调节活性的研究三萜皂苷被认为是一系列具有生物活性的化合物,由植物为阻碍致病菌以及食草动物的侵袭而产生。“皂苷”在这里是指一类结构上具有一个糖苷配基,一个皂苷配基共价结合在一个或多个糖部分的天然化合物。大多数己知的三萜皂苷都属于植物来源的次级代谢物,尽管有少部分的皂苷也在海洋动物中被发现,诸如海参和海星等。合成三萜皂苷的能力广泛分布于被子植物门的植物,无论是双子叶植物还是单子叶植物。天然来源三萜皂苷的研究已取得相当大的进展,通常基于它们的苷元骨架结构来进行分类,例如：齐墩果烷型,乌苏烷型,羽扇烷型,何帕烷型,环菠萝烷型,甘遂烷型,羊毛脂甾烷型,达玛烷型,葫芦烷型以及海参烷型等等。三萜皂苷的生物和药理活性最近在很多研究中被证实,例如：抗过敏反应,抗动脉粥样硬化,抗血栓,抗菌,抗补体,抗糖尿病,避孕,抗真菌,抗炎,抗利什曼虫,抗疟原虫,抗肥胖,抗增殖,治牛皮癣,止痉挛,抗病毒,细胞毒性,抗肿瘤,解毒作用,胃保护,溶血性,肝脏保护,免疫调节,抗酶,抗骨质疏松,杀虫剂,胰岛素样,膜多孔性,灭螺的,神经保护,抗内皮细胞功能紊乱以及蛇毒解药等等。由于三萜皂苷是某些中医药中的活性成分并且具有较高价值的药理学特性,在药物开发领域得到研究人员越来越多的重视。因此,大量的工作已经投入到研究三萜皂苷的作用机制之中,同时探究它们的工业化潜力。尽管在广泛分布的植物中都有产生三萜皂苷能力,但是其构效关系的理解仍没有在哪怕一种植物中被完全阐明。在本工作中,我们集中于三萜皂苷的免疫调节活性的研究,并且运用它们的化学特征和机器学习算法发展了预测模型。建立模型所需的数据集由现存文献中手工检索而获得。该机器学习模型以预测三萜皂苷的免疫调节活性而首次建立。很明显的,本研究中的方法可能会对三萜皂苷构效关系的研究和理解提供帮助。本研究中的结果会帮助研究人员预测皂苷活性以及揭示其药理作用的机制。
[Abstract]:Natural products play an important role in cancer chemotherapy, immunomodulation, neuroprotection, regulation of energy metabolism, anti microbial and cardiovascular disease treatment. They are likely to provide us with the structure of many pilot compounds and can be used as a template for drug delivery in the prevention and treatment of various diseases. The anti-tumor activity of natural products and the immunomodulatory activity of three terpenoid saponins in natural products are studied. The prediction models are established by machine learning methods and the key chemical structures that can describe the corresponding bioactivity of natural products are found. The methods and results of this paper are based on natural products. The development and utilization of drugs, the treatment of disease, and the study of the structure-activity relationship of natural products. (1) the research method of natural product antitumor research methods usually can obtain some key structural properties, such as some chemical bonds or chemical groups. In the study of tumor activity, we usually first determine the effect of this structural feature on a single cancer type or cell line. In this article, we first developed a machine learning method to comprehensively predict the anti-tumor effects of natural products on a series of cancer cell lines. Here we use the base of the cancer cell line at the same time. The chemical structure features (chemical descriptors) of group information (gene expression) and natural products. When these machine learning models are applied to two datasets, training sets and independent test sets, these prediction models show very good prediction accuracy. In addition, we also prove that this method has two natural products, namely ginger. The ability to predict the antitumor activity of lutein and resveratrol, which indicates that the method can effectively predict the one-to-one correspondence between the two natural products and various cancer cell lines, and also linked the patient's genomic information to the sensitivity of natural products. In recent years, many natural products have been isolated and purified in the laboratory and have been proved to have some antitumor activity, such as camptothecin, Changchun anthraquinone, maxoquinone, paclitaxel and so on. Products have played a key role in the treatment of a variety of cancers such as melanoma, leukemia, and breast cancer. In addition, more and more new derivatives based on natural product structures have become antitumor candidates. However, it is extremely time-consuming and laborious to screen natural products by experimental methods. Methods to study its structure-activity relationship is a good choice. The traditional methods for studying structure-activity relationships are mostly chemical background researchers based on the basic skeleton of pilot compounds to synthesize a series of analogues and predict their anti-tumor activity according to experience. Using structural information to try to predict the inhibition of a single cancer cell line or a single tissue type. Although many results have been achieved, the problem is far from being well solved. In order to obtain a comprehensive analysis, we intend to use a new approach based on chemical structure and genomic information. A simultaneous detection of the sensitivity of several known cancer cell lines to hundreds of natural products. We first established a machine learning method to predict the response of cancer cell lines to natural products by using the gene expression data of the cancer cell lines and the chemical descriptors of the natural product structure. The experimental results of the training set and test set have proved that Our approach can be more satisfactory in predicting the sensitivity of hundreds of cancer cell lines. It also demonstrates the combination of structural features and gene expression characteristics, which has an important role in determining the antitumor activity of natural products, the project's prediction of drug reactions and the discovery of new natural sources. (2) study on the immunomodulatory activity of three terpene saponins, three terpenoid saponins are considered to be a series of bioactive compounds, which are produced by plants to impede the invasion of pathogenic bacteria and herbivorous animals. "Saponins" here refer to a class of glycosides with a glycoside ligand and a covalent binding of saponins. Natural compounds in one or more sugar parts. Most known three terpenoid saponins belong to the secondary metabolites of plant origin, although a few of the saponins are found in marine animals, such as sea cucumbers and starfish. The ability to synthesize three terpenoid saponins is widely distributed in plants of the quilt gate, whether dicotyledonous or single. Cotyledon. The study of three terpenoid saponins from natural sources has made considerable progress, usually based on their glycoside skeleton structure, such as oleanane type, uranane type, lupane type, lehpane type, ring bromelane type, calamane type, lanolin sterane type, damanane type, cucurbanane type and sea cucumber, and so on. Three terpenes. Biological and pharmacological activities of saponins have recently been confirmed in many studies, such as antiallergic reaction, anti atherosclerosis, antithrombotic, antithrombotic, antidiabetic, antidiabetic, contraceptive, antifungal, anti inflammatory, anti Leishmania, antimalarial, anti obesity, anti proliferation, treatment of psoriasis, antispasm, antiviral, cytotoxicity, antitumor, detoxification, antidote, antidote, and gastric preservation. Protection, hemolysis, liver protection, immunomodulation, anti enzyme, anti osteoporosis, insecticide, insulin like, membrane porous, snail, neuroprotection, anti endothelial cell dysfunction and snake venom antidote and so on. Three terpenoid saponins are active components in some Chinese medicine and have higher pharmacological properties and are obtained in the field of drug development. Researchers have paid more and more attention. So a lot of work has been put into the study of the mechanism of three terpenoid saponins and their industrialization potential. Although the ability to produce three terpenoid saponins in widely distributed plants, the understanding of its structure-activity relationship is still not fully elucidated in any kind of plant. In our work, we focused on the study of the immunomodulatory activity of three terpenoid saponins and developed a prediction model using their chemical characteristics and machine learning algorithms. The data set required for the establishment of the model was manually retrieved from existing literature. The machine learning model was established for the first time to predict the immunomodulatory activity of three terpenoid saponins. The method in this study may be helpful to the research and understanding of the structure-activity relationship of three terpenoid saponins. The results in this study will help researchers to predict the activity of saponins and reveal the mechanism of their pharmacological effects.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：R96

【相似文献】