CAREER: Building an informative machine learning model of plant gene regulatory control

尽管植物研究在可持续食品和能源生产、对环境负责的材料生物合成以及安全有效的药物等方面具有巨大的变革潜力，但对植物基因如何“开启”或“关闭”以生产这些有用产品的基本理解仍然是一个主要的科学瓶颈。为了了解这些遗传途径是如何发挥作用的，我们必须识别和理解最终控制这些基因的DNA元件的模式。为了解决生物信息学中的这一基本科学挑战，该项目将使用模式识别技术，以详细了解DNA编码信息控制植物基因的方式。为了做到这一点，我们将开发特定的计算技术来构建和解释模式识别模型。这项研究最重要的方面是创造具有识别能力的算法，当一些不同的DNA元素模式为基因控制提供生物相关的解决方案时。该项目开发的算法、模型和生物学成果将提供给科学界，包括植物科学界。基本的计算方法预计将适用于许多其他科学和工程领域。该项目开发的教学和培训模块将包括扩展到我们俄勒冈州农村地区服务不足的社区，包括为高中生提供实践科学体验，以及教师培训。这些模块将特别引起人们对植物科学中受过计算机培训的研究人员的需要，以及对植物基因调控网络作为新药发现、农业和材料生物合成挑战的基础的极大提高的理解的需要。转录因子对基因表达的组合调控是从植物到人类的真核生物生命树上最基本的调控机制之一。基因起始区周围的DNA区域称为启动子区域，包含称为转录因子结合位点(TFBSS)的短调控序列元件。在植物中，尽管有许多令人兴奋的潜在进展取决于对植物基因调控的详细了解，但对导致基因表达的特定TFBS模式知之甚少。该项目使用并扩展了由Megraw实验室开发的机器学习模型，该模型能够高精度和高分辨率地预测转录起始位点(TSS)的存在。这个模型的一个主要优点是，它提出了一组特定的TFBS：启动子之间的相互作用，有可能‘打开’特定的基因。该模型新颖地利用了目前可用的高通量TSS数据来检查模式植物拟南芥中启动子的全球结构，并在初步研究中表明，植物和动物DNA调控代码之间存在显著的差异。这个项目的目标是严格测试和扩展这一有前景的新计算方法，以解剖生物样本中哪些TFBS集合是基因上调的最佳预测因子。拟南芥基因组的相对简单使这一挑战变得容易处理，但能够为多细胞生物体中的基因调控提供变革性的洞察。该项目的结果可以在http://megraw.cgrb.oregonstate.edu/projects/PlantRegThis上找到，该奖项反映了国家科学基金会的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Despite the enormous potential of plant research for transformative advances in sustainable food and energy production, environmentally responsible materials biosynthesis, and safe effective medicines, a basic understanding of how plant genes are 'turned on' or 'turned off' in order to produce these useful products remains a primary scientific bottleneck. In order to understand how these genetic pathways function, one must identify and understand the patterns of DNA elements that ultimately control these genes. To address this fundamental scientific challenge in biological informatics, this project will use pattern recognition techniques in order to develop a detailed understanding of the way in which plant genes are controlled by DNA-encoded information. In order to do this, we will develop specific computational techniques for building and interpreting pattern recognition models. The most important aspect of this research is creating algorithms with the ability to recognize when a number of different DNA element patterns provide biologically relevant solutions for gene control. The algorithms, models, and biological outcomes developed by this project will be made available to the scientific community, including the plant science community. The underlying computational methods are expected to be applicable in many other fields of science and engineering. Teaching and training modules developed by this project will include outreach to our underserved communities of rural Oregon, including hands-on science experiences for high school students, as well as teacher training. The modules will draw special attention to the need for computationally trained researchers in the plant sciences, and to the need for a dramatically increased understanding of plant gene regulatory networks as a foundation for new drug discovery, agricultural, and materials biosynthesis challenges. The combinatorial control of gene expression by Transcription Factors (TFs) is one of the most fundamental regulatory mechanisms across the eukaryotic tree of life, from plants to humans. The DNA region surrounding the start of a gene, called the promoter region, contains short regulatory sequence elements known as Transcription Factor Binding Sites (TFBSs). In plants, despite the many exciting potential advances that hinge on a detailed understanding of plant gene regulation, relatively little is known about the specific TFBS patterns that lead to gene expression. This project uses and extends a machine learning model developed by the Megraw lab that is able to predict the presence of Transcription Start Sites (TSSs) with high accuracy and resolution. A primary advantage of this model is that it suggests specific sets of TFBS:promoter interactions which have the potential to 'turn on' a particular gene. This model makes novel use of currently available high-throughput TSS data to examine the global structure of promoters in the model plant Arabidopsis, and in preliminary research suggests striking unanticipated differences between plant and animal DNA regulatory codes. The goal of this project is to rigorously test and expand this promising new computational approach for dissecting which TFBS sets are optimal predictors of gene up-regulation in a biological sample. The relative simplicity of the Arabidopsis genome makes this challenge tractable, yet capable of providing transformative insight into gene regulation in a multicellular organism. The results of this project can be found at http://megraw.cgrb.oregonstate.edu/projects/PlantRegThis award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.