AI4PhotMod - Artificial Intelligence for parameter inference in Photosynthesis Models

光合作用将大气中的二氧化碳固定下来，推动作物和自然植被的生长，从而提供可再生的食物、燃料、药品和纤维。提高光合效率也越来越被认为是提高作物性能的一种策略。通过测量植物与周围空气之间的二氧化碳和水蒸气交换，可以确定光合作用过程中吸收了多少碳，平行蒸腾了多少水，以及这些通量如何随着环境条件的变化而变化，无论是生长过程中的短期变化，还是气候变化导致的长期变化。为了分析这些气体交换数据，科学家们使用了非常简单的模型，只对光合作用中涉及的生化过程进行了基本的描述，尽管对光合作用和二氧化碳同化反应的代谢网络有更详细的了解，而且存在详细的计算机模型，比目前使用的简单模型包含了更多的这方面的知识。在专注于提高光合作用效率的工作中，使用过于简单的模型是有问题的，因为它们不包含所涉及过程的足够详细的表示，因此不能可靠地为工程策略的设计提供信息。然而，更合适的详细模型的相当复杂性导致了一个主要的参数化问题。由于模型校准数据的缺乏，在有数据的情况下，基于经典方法从这些数据中估计参数需要很长时间。这项提案将解决这两个问题。使用人工智能方法，我们将开发一种参数预测算法，一旦训练完毕，只需几分钟即可运行。我们将在已经被广泛采用和易于使用的标准化协议生成的最小数据集上开发这种方法。这项工作的结果将允许在大量已有数据以及广泛的新研究项目中应用最先进的光合作用模型。

Photosynthesis fixes carbon dioxide from the atmosphere to drive growth of crops and natural vegetation, thus providing renewable supplies of food, fuel, medicine and fibre. Improving photosynthetic efficiency is also increasingly being recognised as a strategy to enhance crop performance. Measurements of the exchange of carbon dioxide and water vapour between plants and the air surrounding them are used to determine how much carbon is assimilated during photosynthesis, how much water is transpired in parallel, and how these fluxes may change with a change in environmental conditions, either short-term during growth, or long-term due to climate change.To analyse these gas exchange data, scientists use very simple models with only a basic representation of the biochemical processes involved in photosynthesis, despite the fact that much more detailed understanding of the metabolic network of reactions involved in photosynthesis and CO2 assimilation is available and detailed computer models exist that incorporate much more of this knowledge than the simple models currently used. The use of overly simple models is problematic in work focused on improving the efficiency of photosynthesis, since they do not contain sufficiently detailed representation of the processes involved and therefore cannot reliably inform the design of engineering strategies.However, the considerable complexity of more appropriate detailed models has led to a major parameterization problem. There is a shortage of model calibration data and where data is available, parameter estimation from this data based on classical methodology takes a very long time. This proposal will address both of these issues. Using an artificial intelligence approach we will develop a parameter prediction algorithm which, once trained, will take only a few minutes to run. We will develop this method on a minimal set of data generated with standardized protocols that are already widely adopted and easy to use. The outcomes of the work will allow application of state of the art models of photosynthesis across a wealth of pre-existing data, as well as a wide range of new research projects.