Collaborative Research: IIBR Informatics: Keeping up with the genomes - Continual Learning of Metagenomic Data

微生物群是地球上随处可见的微生物群落，对帮助消化肠道中的食物很重要。在肠道中，它们可以产生维生素（好的）或毒素（坏的），所以我们需要了解这些微观群落中存在什么生物体和基因。该项目使用人工智能（AI）来识别生活在微生物组中的生物体及其基因。由于数据量的快速增长，通常需要在获得新数据时反复重新分析，因此现有的工作受到了阻碍。这样的过程不仅效率低下，而且越来越不可持续，即使对于我们不断增长的计算资源也是如此。这种方法是独特的，因为它使用较少的计算能力。该系统无需不断重新输入大量数据，而是能够在不需要重新输入或重新分析先前数据的情况下回忆和重用先前的信息，从而节省了大量的计算时间和资金。我们的目标是找到在不牺牲准确性的情况下实现最佳成本节约的人工智能方法。在微生物组实验中也发现了许多未识别的生物，并且被丢弃，并且从未在其他实验中用于识别相同的生物。一种基于人工智能的方法将保留、记住和重用它们的信息，以防这些新的生物体在以后的其他实验中再次出现，并最终帮助识别它们。该方法可在将来对生物进行识别时，自动更新旧数据和知识库。该项目将开发一个动态的、可扩展的、半监督的学习框架，该框架将使用大量未标记的实验数据不断更新分类模型。除了创建可以利用参考和实验数据的更丰富的模型外，主要的创新是该模型将识别未知的生物体和蛋白质，并将其整合到参考数据库中，以便将来更新模型。该框架将在每年提交给微生物组计算网站MG-RAST的数百个宏基因组研究（可能由数千个样本组成）上进行验证。科学家们使用MG-RAST上传他们的微生物组，以研究和改进农业、诊断、医药、制造生物燃料以及微生物对其有深刻影响的各种其他应用。这项工作将有助于大学生对人工智能及其在微生物组中的应用进行培训。研究结果将通过夏季研讨会与其他教育工作者和研究人员广泛分享。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Microbiomes are communities of microscopic organisms that are found everywhere on earth and are important in help to digest food in the gut. In the intestines, they can produce vitamins (good) or toxins (bad), so we need to understand what organisms and genes are present in these microscopic communities. This project uses artificial intelligence (AI) to identify organisms and their genes that live in microbiomes. Existing works for this effort have been hampered due to very rapidly growing amount of data, which often need to be repeatedly re-analyzed as new data become available. Such a process is not only inefficient, but is increasingly unsustainable, even for our growing computational resources. This approach is unique because it uses less computing power. Instead of continuously reentering massive amounts of data, the proposed state-of-the-art system has the ability to recall and reuse prior information without requiring reentering or re-analyzing prior data,saving substantial computing time and ultimately money. The goal is to find AI methods that achieve the best cost savings while not sacrificing accuracy. Many unidentified organisms are also found in microbiome experiments and are discarded and never used to identify the same organisms in other experiments. An AI based approach will keep, remember, and reuse their information in case those new organisms show up in again later in other experiments and eventually help in their identification. If the organism is identified in the future, the method can automatically update old data and the knowledgebase effectively and efficiently.This project will develop a dynamic, scalable, and semi-supervised learning framework that continually updates a classification model, with large unlabeled, experimental data. In addition to creating richer models that can leverage both reference and experimental data, the primary innovation is that the model will identify unknown organisms and proteins and integrate them into reference database for future model updates. This framework will be validated on the hundreds of metagenomic studies (composed of potentially thousands of samples) annually submitted to the microbiome computing website MG-RAST. MG-RAST is used by scientists to upload their microbiomes to study and improve agriculture, diagnoses, medicine, making biofuels, and a variety of other applications on which microorganisms have a deep effect. This work will contribute to college student training on artificial intelligence and its application to the microbiome. Results will be shared broadly with other educators and researchers through summer workshops.This 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.