Using machine learning to tell the time

该项目旨在采用机器学习方法来准确预测植物的复杂性状。生物学家测量的许多表型和性状都很复杂，耗时且需要特定的技能。解决这个问题的一个强大方法是开发健壮的代理度量。该项目最初将开发一个模型，利用转录组学数据预测内部生物时间，以确定一个代理基因集，该基因集的表达可以准确预测采集样本的生物时间。在过去的10年里，转录组学分析已经成为一种简单、可靠、相对便宜的分析方法。代理基因集将用于公共转录组数据集，以研究基因型和环境如何影响生物时间。这项工作将以今年发表的一篇论文为基础（Gardiner et al. 2021），该论文应用机器学习来预测模式植物拟南芥（Arabidopsis thaliana）中复杂的时间昼夜节律基因表达模式。该项目将继续开发其他重要性状的代理基因集，如作物产量、抗逆性、抗病性和氮利用效率。

The project aims to take a machine learning approach to accurately predict complex traits in plants. Many of the phenotypes and traits that biologists measure are complex, time consuming and require a specific skill set. One powerful approach to getting around this is to develop robust proxy measurements. The project will initially develop a model to predict internal biological time using transcriptomic data, to identify a proxy gene set for which the expression can accurately predict the biological time the sample was taken. Over the last 10 years transcriptomic analysis has become a simple, robust, and relatively cheap assay. The proxy gene set will be used across public transcriptomic data sets to investigate how genotype and environment affect biological time. The work will build on a paper published this year (Gardiner et al. 2021), which applied machine learning to predict complex temporal circadian gene expression patterns in the model plant Arabidopsis thaliana. The project will go on to develop proxy gene sets for other important traits, such as crop yield, resilience, disease resistance and nitrogen use efficiency.