植物高通量基因型和表型数据计算分析及工具开发

发布时间：2018-10-11 09:02

【摘要】：组学方法包括基因组学、蛋白组学、转录组学、表观基因组学和代谢组学。整合表现型分析方法辅助作物育种,形成了新的组学方法-表型组学(包括对组织的生物物理和生物化学特性的高通量分析)。基因组学和表型组学是生物科学的两个重要分支,它们处于多组学家族的两端。如今生物学的一个中心目标是在基因组和表型组之间建立完整地功能连接,我们称之为基因型表现型图谱。细胞系统是涉及到成千上万个基因转录调节的基因表达的产物。因此,阐明转录调控网络的机理,不仅对于解细胞工作机理是必须步骤,而且对于发现新的生物分子靶标也是十分有利的。在基因组学研究中,从表达数据来预测基因调控网络是一个具有挑战性的工作。目前,已有许多方法(从监督学习到非监督学习)被开发出来以应对这个挑战。其中最具有前景的是基于支持向量机的方法(SVM)。我们需要对使用在不同的生物实验条件和网络规模下、基于不同核心的监督学习SVM方法进行的全面的分析来比较其预测的准确性。因此,基于SVM,我们开发了一个叫(CompareSVM)的工具来比较不同基因调控网络的推理方法。通过CompareSVM,我们使用不同的SVM内核方法对不同大小基因芯片以及二代测序的数据集进行细节上的模拟。从CompareSVM反馈的结果表明,推理方法的准确性取决于实验条件与网络尺寸的性质。植物表型研究的局限性限制了我们对数量形状遗传的剖析,尤其是那些涉及产量和抗逆性(例如,增产潜力,提高抗早,耐热性和养分效率等)的遗传效应。如今,开发有效的高通量表型分析平台虽然仍处在瓶颈期。然而,生物、传感器以及高性能计算的进展正在为此铺平道路。高通量表型分析是剖析植物表型组分的重要技术,为了量化植物生长以及基于表型性状等分析,有效的图像处理性能和特征提取是分析中必不可少的。所以,对于各种不同的植物物种,基于实时收集不同范围的图像数据,有必要开发一个支持传输来自不同采集环境的图像并能大规模进行图像分析的系统。目前,已经开发了一个非破坏性方式捕获植物广泛和深入的表型数据的高通量分型平台,而这些进展推进了我们对植物生长以及植物响应气候和环境改变的见解。基于这些发展,越来越高效的作物遗传改良以满足子孙后代的需要。在植物表型分析中,以一种无损方式来对植物表型进行参数评估的数字图像分析是一项十分重要的工作。如今,用于图片分析的一些基于不同的要求的筛选系统已经被开发出来,并且部分可以商业使用。在表型组学的研究中,植物器官尤其是独立叶片的分割和鉴别是基于图片植物表型分析的最大难点之一。全自动化的表型分析系统虽然可以持续的收集植物图片,但是同时也带来了高劳动力,高成本和高维护代价等问题。所以我们需要一个更灵活的系统来适应不同的植物背景、植物照明以及相似变化的问题。因此,我们通过扩展imageJ的艺术照片处理算法库,开发了一个可以完全免费获取的ImageJ的插件-HTPPA,与HTPPA同时开发的还有一些可以综合探索高通量表型分析的实用工具。通过利用植物结构和形态特征提高植物和独立叶片的分割,我们可以推进大范围高通量的表型分析并建立基因型和表型之间的联系。基因组学和表型组学是生物科学的两个最重要基本分支,是多组学的两个端点。科学技术的进步增加了可用的多组学数据的广度,从全基因测序数据到广泛的转录组、甲基化组和代谢组数据。关键的目标是通过定义用于预测表型性状和结果的有效的模型,建立基因型和表型之间的全面的功能连接。利用全基因组关联分析(GWAS)方法处理高通量和高维度的基因型和表型数据,可以从中发现基因型和表型存在的灰色区域。GWAS及其类似方法的应用和多组学数据的整合开始发现基因型的变异对表型多样性的贡献。通过使用系统生物的方法整合广泛水平的组学数据是至关重要的,这样可以进一步弥合基因组学和表型组学,最终可以基于基因型的贡献做出表型的准确
[Abstract]:Group studies include genomics, proteomics, transcriptome, epigenomics, and metabolism. The entire phenotype analysis method assisted crop breeding, forming a new group learning method-Phenotypic group study (including high-throughput analysis of the biophysical and biochemical characteristics of the tissue). Genomics and phenotypic groups are two important branches of the study, which are at the ends of multiple groups of studies. One central goal of biology today is to establish a complete functional connection between the genome and the phenotypic group, which we call genotype phenotypes. Cell systems are products involved in the expression of genes involved in the transcription regulation of tens of thousands of genes. Therefore, it is necessary to clarify the mechanism of transcriptional regulation network, not only to solve the mechanism of cell work but also to find new targets for biological molecules. In genomics research, it is challenging to predict gene regulation networks from expression data. At present, many methods have been developed (from supervisory learning to non-supervised learning) to address this challenge. wherein the most promising is a support vector machine-based method (SVM). We need to compare their predicted accuracy with a comprehensive analysis using different core-based supervised learning SVM methods under different biological experimental conditions and network sizes. Therefore, based on SVM, we developed a method called CompareSVM to compare the reasoning methods of different gene regulation networks. Through the CompareSVM, we use different SVM kernel methods to simulate different size gene chips and second-generation sequencing data sets. The results of the feedback from the CompareSVM show that the accuracy of the reasoning method depends on the experimental conditions and the nature of the network size. The limitations of plant phenotype studies have limited our analysis of quantitative shape inheritance, especially those related to yield and stress resistance (e.g., increased yield potential, improved early resistance, heat resistance, and nutrient efficiency, etc.). Nowadays, the development of effective high-throughput phenotype analysis platform is still in bottleneck period. Progress in biology, sensors, and high-performance computing, however, is paving the way for this. High-throughput phenotype analysis is an important technique to analyze the phenotypic components of plants. In order to quantify plant growth and phenotypic traits, effective image processing performance and feature extraction are essential in the analysis. Therefore, for a variety of different plant species, based on the real-time collection of different ranges of image data, it is necessary to develop a system that supports transmission of images from different acquisition environments and can perform image analysis on a large scale. At present, a high-throughput typing platform that captures widespread and in-depth phenotypic data of plants has been developed, which advances our insights into plant growth and plant response climate and environmental changes. Based on these developments, more and more efficient crop genetic improvements meet the needs of future generations. In plant phenotypic analysis, digital image analysis of parametric evaluation of plant phenotypes in a non-destructive manner is a very important task. some screening systems based on different requirements for picture analysis are now developed and partly commercially available. In the study of phenotypic groups, the segmentation and identification of plant organs, especially the independent leaves, is one of the greatest difficulties based on the phenotypic analysis of picture plants. The full-automatic phenotypic analysis system can collect plant pictures continuously, but also brings problems such as high labor force, high cost and high maintenance cost. So we need a more flexible system to adapt to different plant backgrounds, plant lighting, and similar changes. As a result, we have developed an ImageJ plug-in-HTPPA that can be acquired free of charge by expanding the art photo processing algorithm library of imageJ, and a number of utilities that can be developed simultaneously with the HTPPA to explore a high-throughput phenotype analysis. By improving the segmentation of plants and individual leaves using plant structures and morphological features, we can advance large-range high-throughput phenotype analysis and establish linkages between genotypes and phenotypes. Genomics and Phenotypic group studies are two of the most important basic branches of science and technology, and are two endpoints of multi-group learning. Advances in science and technology have increased the breadth of available sets of learning data, from full-gene sequencing data to a wide range of transcription groups, methylation groups, and metabolic group data. The key objective is to establish a comprehensive functional connection between the genotype and phenotype by defining an effective model for predicting phenotypic traits and results. Genetic and phenotypic data from high flux and high dimensions can be processed using the whole genome association analysis (GWAS) method, and a gray area exists from which the genotype and phenotype can be found. The application of GWAS and its similar methods and the integration of multiple sets of learning data began to find the contribution of genotypic variations to phenotypic diversity. It is of vital importance to integrate a wide range of group learning data through the use of a system biology approach, which can further bridge genomics and phenotypic groups and ultimately make the phenotype accurate based on the contribution of the genotype
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：Q943.2;Q811.4

【相似文献】