ABI Innovation: Gini-based methodologies to enhance network-scale transcriptome analysis in plants

在生物学中，网络技术已被应用于解释基因之间的相互作用，包括蛋白质的物理相互作用和转录因子与靶标之间的调控关系。尽管已经开发了许多方法来从表达数据推断网络，但仍有几个计算挑战未解决，例如，如何导出转录因子和靶标之间的非线性关系，如何将网络适当地分解为各个子网络模块，如何通过网络规模比较预测生物学上重要的基因，如何整合和使用生物相互作用数据的异构形式，以促进网络分析，以及如何无缝地可视化一个大规模的网络交互式数据挖掘。为了解决这些问题，这个项目的主要目标是开发一个软件包-基尼网络分析工具包（GNAT），利用基尼为基础的方法：一个家庭的数学解决方案，已被广泛用于经济学，物理学，信息网络和社交网络在分析非正态分布的数据。GNAT中的核心功能模块和算法包括使用监督机器学习方法来推断转录网络，Gini相关系数来推导非线性调控关系，Gini回归分析来分解时间序列网络，Gini指数来测量和比较不同生物条件下模块和基因的网络属性分布，并最终通过系统扰动和决策树分析发现生物学上重要的基因。首席研究员亦会开发一个名为BioNetscape的网络浏览器，利用k核心分解算法、Ajax技术和GPU（图形处理器）计算技术，有效地组织和可视化GNAT产生的大量网络数据。GNAT将以R语言实现，并以简化的工作流程组织，以弥补传统基因级转录组分析方法的缺点。GNAT软件将大大促进正在进行的植物研究网络开发项目。GNAT将被集成到iPlant Discovery Environment、The Arabidopsis Information Resource（TAIR）、Plant Expression Database（PLEXdb）和其他联盟数据库中，以增强植物网络分析和基因发现的功能。GNAT还将被纳入银河和GenePattern平台，以提供方便用户的图形界面。源代码和R软件包将被发布到公共领域，以便在植物，动物和微生物生物学中更广泛地使用。为了将研究融入教育，PI？的实验室将开发一个基于网络的虚拟下一代测序研讨会，用于培训非生物信息学专家的生物学家分析基因组、表观基因组、转录组和小RNA数据。工作坊的课件由PI课堂上准备的教学材料、自我练习数据集和虚拟UNIX网络控制台组成，用于培训生物学家分析不同类型的下一代测序数据，对编程技能的要求最低。该项目明确解决了多层次的跨学科研究培训，这将鼓励亚利桑那大学计算机科学和数学领域代表性不足的群体参与，他们将致力于回答生物学问题。来自亚利桑那大学ASEMS（亚利桑那科学、工程和数学学者）和IGERT项目的学生将参加PI的团队，开发GNAT、BioNetscape和VNW，并在他们的研究中使用这些工具。

In biology, network techniques have been applied to interpret the interactions between genes, including the physical interactions of proteins and regulatory relationships between transcription factors and targets. Although numerous methods have been developed to infer a network from expression data, several computational challenges remain unsolved, such as, how to derive non-linear relationships between transcription factors and targets, how to properly decompose a network into individual sub-network modules, how to predict biologically significant genes via network-scale comparisons, how to integrate and use the heterogeneous forms of biological interaction data to facilitate network analysis, and how to seamlessly visualize a large-scale network for interactive data mining. To solve these problems, the primary goal of this project is to develop a software package - the Gini Network Analysis Toolkit (GNAT) that utilizes the Gini-based methodologies: a family of mathematical solutions that have been widely used in economics, physics, informatic networks, and social networks in analyzing non-normally distributed data. The core functional modules and algorithms in the GNAT include the use of supervised machine learning methods to infer transcriptional networks, the Gini correlation coefficient to derive non-linear regulatory relationships, the Gini regression analysis to decompose a time-series network, the Gini index to measure and compare the distributions of the network properties of modules and genes under different biological conditions, and eventually the discovery of biologically important genes with system perturbation and decision tree analysis. The PI will also develop a network explorer, BioNetscape, to efficiently organize and visualize the tremendous amount of network data generated from the GNAT using the k-core decomposition algorithm, Ajax technology and GPU (graphical processing unit) computing techniques. The GNAT will be implemented in R and organized as a streamlined workflow to compensate the shortcomings of the traditional gene-scale transcriptome analysis methods.The GNAT software will greatly facilitate the ongoing network development projects in plant research. The GNAT will be made available to be integrated into the iPlant Discovery Environment, The Arabidopsis Information Resource (TAIR), Plant Expression Database (PLEXdb) and other consortium databases to enhance the function of network analysis and gene discovery in plants. The GNAT will also be integrated into the Galaxy and GenePattern platforms to provide a user-friendly graphical interface. The source-code and R packages will be released into the public domain for broader use in plant, animal and microbial biology. To integrate research into education, the PI?s laboratory will develop a web-based Virtual Next Generation Sequencing Workshop for training biologists who are not specialists in bioinformatics to analyze genomic, epigenomic, transcriptomic and small RNA data. The workshop courseware is composed of teaching materials prepared in the PI's class, self-practice datasets and a virtual UNIX web-console for training biologists to analyze different types of next generation sequencing data with minimal requirements for programming skills. This project explicitly addresses cross-disciplinary research training at multiple levels that will encourage the participation of underrepresented groups in computer sciences and mathematics at the University of Arizona, who will work to answering biological questions. The students from the ASEMS (Arizona Science, Engineering, and Math Scholars) and IGERT programs at the University of Arizona will participate in the PI's team to develop the GNAT, BioNetscape and VNW, and use these tools in their research.