RAPID: Machine Learning Methods to Understand, Predict and Reduce the Spread of COVID-19 in Small Communities

持续的COVID-19疫情最近已达到全球大流行状态，并在全球蔓延。疫情的严重性，加上沿着对世界经济和社会造成的巨大影响，迫使各国政府采取紧急措施。至关重要的是，要利用来自可靠来源的现有统计数据，以模拟和评估大流行病传播的动态，不仅要更好地了解这种复杂的系统，而且要学习和制定可能的解决方案，以防止当前和/或未来类似疫情的进一步传播。因此，这项致力于开发COVID-19大流行传播数学模型的研究解决了国家的迫切需求。新墨西哥州高地大学（Highlands University）计算机科学、人类学和计算化学领域的师生组成了一个多元化的团队，旨在为描述和预测COVID-19传播的复杂问题找到解决方案。该多学科项目预计将更好地了解导致COVID-19传播的诸多因素之间的相互联系。将在北方新墨西哥州的地区收集统计数据，包括纳瓦霍族的圣胡安县和麦金利县以及纳瓦霍族以外的洛斯阿拉莫斯县。将向新墨西哥州（NM）部落和卫生当局提交对收集的统计数据的分析沿着该项目的社会文化评估。该项目旨在为预测疾病传播提供科学依据，并将考虑与另一波大流行病的可能性有关的情景。参与该项目的学生将获得应用先进机器学习模型和方法对国家健康、经济和社会危机做出快速稳健反应的宝贵经验。在该研究中，机器学习方法将用于分析不同地区的流行病传播情景，并收集表征传播的数据的最重要特征。该研究团队将使用传统的机器学习技术和先进的方法，如人工神经网络，从而开发病毒发病率模型，捕获线性和非线性域中的依赖关系。这项工作将集中在了解疾病传播方面的多种社会经济因素。这个问题可以看作是一个序列建模问题，因此，递归神经网络和基于其递归细胞的更复杂的模型可能是一个有前途的方向。下一步将是为具有不同社会经济背景和种族的小型孤立社区收集数据集-将生活在纳瓦霍保留区的纳瓦霍印第安人与洛斯阿拉莫斯县（NM）进行比较-并测试开发的模型对这些地区的适用性。在排列上可用的时空数据在性质上是异质的。本研究的一个重要目标是根据流行曲线行为的相似性对收集的数据进行分类，然后根据这种分类为不同的地区建立单独的模型。该模型将用于预测未来的事件，并为抑制和预防未来的病毒爆发提供最有效的非医疗建议。这项研究得到了材料研究与教育伙伴关系（PREM）计划和凝聚态物质和材料理论（CMMT）的支持。数学和物理科学理事会材料研究部的项目使用冠状病毒援助、救济、该奖项反映了NSF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。

The ongoing COVID-19 outbreak has recently reached pandemic status spreading all around the world. The severity of the pandemic, along with an enormous impact on world’s economy and society, has forced governments to introduce emergency measures. It is essential to utilize the available statistical data from trusted sources in order to model and evaluate the dynamics of the pandemic spread, to not only better understand such complex systems, but to learn and develop possible solutions to prevent further spread of current and/or similar future outbreaks. Thus, this research, devoted to the development of mathematical models of COVID-19 pandemic spread, addresses an urgent national need. Faculty and students in computer science, anthropology, and computational chemistry at New Mexico Highlands University have formed a diverse group for finding a solution to the complicated problems of the description and prediction of COVID-19 spread. This multidisciplinary project is expected to yield a better understanding of the interconnections among many factors that contribute to the spread of COVID-19. Statistical data will be collected in regions of Northern New Mexico, including San Juan and McKinley Counties in the Navajo Nation and Los Alamos county outside of the Navajo Nation. Analysis of the collected statistical data along with socio-cultural assessment from this project will be presented to New Mexico (NM) tribal and health authorities. The project will aim to provide a scientific basis for the prediction of disease spread and will consider scenarios associated with the possibility of another wave of the pandemic. Students from this minority-serving institution involved in the project will obtain valuable experience in the application of advanced machine learning models and methods in providing fast robust reaction to a national health, economic, and societal crisis.In this study, machine learning methods will be used to analyze pandemic spread scenarios in different regions and to glean the most important features of the data characterizing the spread. The research team will use both traditional machine learning techniques and advanced methods, such as artificial neural networks, allowing development of virus incidence model capturing dependencies in both linear and nonlinear domains. The work will concentrate on understanding disease spread with regard to multiple socioeconomic factors. The problem can be treated as a sequence modeling one; so, recurrent neural networks and more complex models based on their recurrent cells might be one promising direction. The next step will be to assemble datasets for small isolated communities with different socioeconomic backgrounds and ethnicities – comparing Navajo Indians living on the Navajo reservation to Los Alamos County (NM) – and to test the applicability of the developed model to these regions. The spatiotemporal data available on the spread is heterogeneous in character. An important goal of this research is to classify the collected data with respect to the similarity in the epidemic curve behavior and then build separate models for different regions according to this classification. The proposed model will be used for prediction of future incidents and to produce the most effective non-medical recommendations for suppression and prevention of future viral outbreaks.This research is supported by the Partnerships for Research and Education in Materials (PREM) program and the Condensed Matter and Materials Theory (CMMT) program in the Division of Materials Research in the Directorate for Mathematical and Physical Science using supplemental funds made available by the Coronavirus Aid, Relief, and Economic Security (CARES) Act.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.