Modern statistical techniques for assessing and predicting herbicide performance

该项目将为图形数据开发现代回归方法。随着用于化学测试的筛选工具的激增，现在可以轻松创建大量的化学品数据库。另一方面，可用于分析这些大型数据库的严格统计方法的开发尚处于起步阶段，进一步开发以促进化学发现势在必行。在这个项目中，我将开发统计模型和方法来评估化合物的描述性特征和筛选测试的性能，并相应地计算每种化学品的定量得分。我将应用这种方法来解决先正达提供的真实的问题，使用的数据来自他们对除草剂的筛选实验。通常情况下，数千种潜在的除草剂将在实验室中进行一系列筛选测试（测定测试），每次无效的化合物将被丢弃，其余的将根据一套更复杂的标准进行评估，最后几种将进行严格的田间试验。显然，早期试验的数据将表现出高度的不确定性和主观性。与先正达一起确定了该项目不同可能发展的动机和出发点。在项目的开始阶段，我将开发一个模型来预测除草剂在每次测试中的性能，使用的信息包括剂量、测试的植物种类和化学结构，这些信息可以用图表表示。现代回归方法，如支持向量回归，神经网络和高斯过程回归将被探索。该模型应该利用植物物种和化学品家族之间的关系来提高预测性能潜在的好处是发现新的和有效的除草剂，发展或当前的统计和机器学习技术，为未来的研究和发展提供空间。研究理事会的好处包括通过提供国际上优秀的，以发现为导向的研究，帮助英国成为全球领先的研究国家，指导英国研究界应对全球挑战和政府主导的举措，提供具有经济和社会影响的优秀研究，以满足国家的需求。

This project will develop modern regression methods for graphical data. With the proliferation of screening tools for chemical testing, it is now possible to create vast databases of chemicals easily. On the other hand, the development of rigorous statistical methodology that can be used to analyse these large databases is in its infancy, and further development to facilitate chemical discovery is imperative. In this project, I will develop statistical models and methodology for assessing chemical compounds from their descriptive characteristics and their performance on screening tests, and accordingly compute a quantitative score for each chemical. I will apply this methodology to tackle real problems provided by Syngenta using data from their screening experiments on herbicides. Typically, thousands of potential herbicides will undergo a sequence of screening tests (assay tests) in the lab and each time ineffective compounds will be discarded and the remaining are assessed against a more complex set of criteria, with the final few undergoing rigorous field trials. Evidently, the data from the early trials will exhibit high uncertainty and subjectivity. Motivation and starting points for different possible developments in this project have been identified together with Syngenta. In the starting phase of the project, I will develop a model to predict the herbicide's performance on each test using information such as dosage, plant species tested, and the chemical's structure which can be presented as a graph. Modern regression methods such as support vector regression, neural networks, and Gaussian process regression will be explored. The model should exploit the relationships between plant species and families of chemicals to improve predictive performanceThe potential benefits are discovery of new and effective herbicides, development or current statistical and machine learning techniques, room for future research and development.Benefits to the research council include helping UK become the global leading research nation by offering internationally excellent, discovery-driven research, guiding the UK research community in responding to global challenges and government-led initiatives, deliver excellent research with economic and social impact that addresses the needs of the nation.