基于机器学习优化分子对接筛选肾衰营养胶囊有效成分

发布时间：2018-05-06 20:07

本文选题：慢性肾脏病营养不良 + 中药复方　；参考：《南方医科大学》2017年博士论文

【摘要】：营养不良(PEW)是慢性肾脏病(CKD)患者最常见的并发症之一,是CKD患者预后的独立危险因素。由于其病因复杂,目前缺乏有效的治疗方法。中药组方肾衰营养胶囊治疗CKD-PEW有较好疗效,但确切的机制尚不清楚。由于中药组方较多,成分复杂,难以筛查及诠释其中的有效成分和药理作用机制,不利于中药组方的标准化及优化。如何从大量小分子中筛选真正具有治疗作用的活性成分是目前中医药研究的瓶颈问题。随着近年来生物信息技术的发展,对大数据的分析手段也取得长足进步,其中机器学习广泛运用于各类生物数据处理,是目前最有潜力的生物信息分析手段。机器学习算法可拆解赋权等修饰数据特征,最终获得规律,并利用规律函数对未知数据进行预测。由于中医药组方及机制的复杂性,很难用简易的数学模型描述和预测生物现象并加以调控,而机器学习是预测药物靶点以及判断药效的最优手段。因此以机器学习等生物信息技术为分析手段,为揭示复方中药方剂药效物质与作用机理提供了强有力的方法与技术支撑。从而提取小分子信息,通过对已知药物靶点关系进行训练,可以较好预测靶点的活性成分,因此通过机器学习等手段有望对中药复方作用机制的阐明和优化提供新的研究思路和策略。本课题应用相应数据分析手段对潜在药用靶点及方剂有效成分进行虚拟筛选,并对筛选结果进行体外实验验证。方法:GEO数据库中挑选骨骼肌PEW模型基因表达数据;采用SAM算法筛选差异基因;基于差异基因进行GO富集和KEGG富集预测PEW涉及生物功能机制及参与信号途径;应用MisgDB数据库富集联合启动子富集预测相关转录因子结合motif,从而筛选关键信号和可能药用靶点;从多个中医药数据库中检索组方中的化学成分;应用Vina程序基于自由能对小分子与靶点进行批量对接,预测活性成分;应用四种机器学习模型基于分子指纹,对对接结果进行优化,并最终筛出可能活性成分;应用同位素标记法检测筛选成分对骨骼肌蛋白质代谢率影响,应用WB法检测检测筛选成分对靶点蛋白水平影响;应用报告基因法检测筛选成分对靶点与相关启动子结合影响;应用qPCR法检测筛选成分对关键基因表达的影响。结果显示:骨骼肌PEW涉及到蛋白质分解代谢等多条信号通路,FoxO1可能是其关键靶点。基于FoxO1进行分子对接筛选出17个潜在作用分子,基于指纹谱图应用及其学习模型可将活性预测特异性提升至0.9以上,联合四类机器学习模型决策预测甘草素B(HB)为FoxO1配体。与模型组相比,免疫荧光结果显示HB能显著增加肌管半径(P.05),同位素标记实验显示HB能显著降低PEW模型中蛋白质分解率(P.05),WB结果显示HB能显著抑制FoxO1磷酸化水平(P.05),荧光素酶报告基因结果显示HB能抑制FoxO1结合相应启动子序列,qPCR结果显示HB能显著抑制FoxO1下游萎缩相关基因Atrogin-1及MuRF-1的表达水平。结论:基于分子指纹的机器学习模型能显著提高分子对接的特异性,预测活性成分HB能显著改善骨骼肌PEW状态,其机制涉及抑制FoxO1 Ser256磷酸化及阻断其与相关启动子结合,随之抑制下游萎缩相关基因表达。
[Abstract]:Malnutrition (PEW) is one of the most common complications of patients with chronic kidney disease (CKD) and an independent risk factor for the prognosis of CKD patients. Due to its complicated cause, there is a lack of effective treatment. It has a good effect on the treatment of CKD-PEW, but the exact mechanism is not clear. It is difficult to screen and interpret the effective components and pharmacological mechanisms, which is not conducive to the standardization and optimization of Chinese medicine groups. How to screen real active components from a large number of small molecules is the bottleneck of the research of traditional Chinese medicine. With the development of biological information technology in recent years, the analytical means of large data have also been obtained. The machine learning is widely used in all kinds of biological data processing, which is the most potential means of biological information analysis. The machine learning algorithm can dismantle the embellish data features, and finally obtain the law, and use the law function to predict the unknown data. It is difficult to use the complexity of the traditional Chinese medicine group and mechanism. The easy mathematical model describes and predicts biological phenomena and regulates it, and machine learning is the best means to predict drug targets and to judge the efficacy of drugs. Therefore, biological information technology, such as machine learning, is used as an analytical means to provide a powerful method and technical support for revealing the effective substances and mechanism of compound Chinese medicine prescription. Small molecular information, by training the known drug target relationship, can predict the active components of the target, so it is expected to provide new research ideas and strategies for the elucidation and optimization of the mechanism of the action of Chinese medicine compound by machine learning. This subject applies the corresponding data analysis method to the potential medicinal targets and prescriptions. Virtual screening was carried out, and the screening results were tested in vitro. Methods: GEO database was used to select PEW model gene expression data of skeletal muscle; SAM algorithm was used to screen differential genes; GO enrichment and KEGG enrichment based on differential genes were used to predict the biological function mechanism and participate in signal pathway, and MisgDB database was used to enrich and combine. Promoter enrichment and prediction of related transcription factors combined with motif, thus screening key signals and possible medicinal targets, retrieving chemical components from multiple Chinese medicine databases, using Vina program based on free energy for batch docking of small molecules and targets, predicting active components, and using four machine learning models based on molecular fingerprints, The docking results were optimized and the possible active components were screened out. The effect of screening ingredients on the protein metabolism rate of skeletal muscle was detected by the isotope labeling method. The effects of screening ingredients on target protein level were detected by WB method. The effects of screening ingredients on the target and related promoters were detected by using the report gene method, and the qPCR method was used to detect the target protein. The effects of screening components on the expression of key genes. The results show that skeletal muscle PEW involves several signaling pathways such as protein catabolism and other signaling pathways. FoxO1 may be the key target. 17 potential molecules are screened by molecular docking based on FoxO1, and the specificity of activity prediction can be raised to 0.9 based on the application of fingerprint spectrum and its learning model. Four types of machine learning models were used to predict the glycyrrhizin B (HB) as the FoxO1 ligand. Compared with the model group, the immunofluorescence results showed that HB could significantly increase the radius of the myotube (P.05). The isotope labeling experiment showed that HB could significantly reduce the protein decomposition rate (P.05) in the PEW model, and WB results showed that HB could significantly inhibit FoxO1 phosphorylation level (P.05), fluorescence. The results of the enzyme reporter gene show that HB can inhibit the FoxO1 binding promoter sequence. QPCR results show that HB can significantly inhibit the expression level of Atrogin-1 and MuRF-1 in the downstream atrophy related genes of FoxO1. Conclusion: the machine learning model based on molecular fingerprint can significantly improve the specificity of molecular docking, and it predicts that the active component HB can significantly improve the skeleton. The mechanism of muscle PEW status involves inhibiting FoxO1 Ser256 phosphorylation and blocking its binding with related promoters, thus inhibiting the expression of downstream related genes.

【学位授予单位】：南方医科大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：R286.0

【相似文献】