RAPID: An Ensemble Approach to Combine Predictions from COVID-19 Simulations

决策者和普通公众依靠模型来模拟SARS-CoV-2病毒的传播，并预测感染和死亡的数量。模型预测对于迅速制定缓解新冠肺炎的政策干预措施、预测对卫生保健资源的影响以及制定如何最好地实施和提高公共卫生指南的战略至关重要。现有的新冠肺炎模型产生的预测截然不同，因此造成了人们对它们的使用的困惑和不信任。因此，迫切需要将现有的新冠肺炎模型的广泛范围与它们的预测进行比较。这个项目的目标是在各种模型预测中找到共识，并使不同的模型假设和不确定性变得透明。一个基于网络的交互式仪表盘将作为一个开放和可访问的工具，在现有模型对预测意见一致或不一致的情况下，通知公众、研究人员同行和决策者。由现有模型商定的预测然后可以更有信心和信任地用作有效决策的基础，以拯救生命和资源。除了对实施有效的流行病控制措施和公共卫生战略具有实际重要性外，其他更广泛的影响是为早期职业研究人员提供专业发展机会，为博士后学者提供培训机会。为了在各种模型预测中找到共识，该项目将开发一种新的集成预测方法，该方法(1)对不同的新冠肺炎模拟模型进行比对，(2)使用时间序列聚类技术来统一模型预测。在模型比对阶段，将选择一系列开源和公开可用的新冠肺炎仿真模型，根据它们的参数进行比对，并根据需要运行。将从模型结果中获得一组广泛的预测，其中每个预测代表一个可能的世界，具有相应数量的新病例、死亡和其他感兴趣的数量。时间序列聚类阶段将可能世界投影到特征空间中，并应用聚类算法来寻找相似的可能世界。对于每个集群，将选择一个具有代表性的可能世界，用不确定性度量丰富，并使用仪表板进行可视化。该项目提出了模型对齐方法的理论知识库和用于仿真模型预测的具有代表性的不确定性聚类。将模型预测统一成科学共识可以用来更好地向决策者提供信息，以便他们能够制定拯救生命的干预措施。这一快速奖项由环境生物学部门的传染病生态学和进化项目利用冠状病毒援助、救济和经济安全法案(CARE)的资金颁发。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

Decision-makers and the general public rely on models to simulate the SARS-CoV-2 virus spread and predict the number of infections and fatalities. Model predictions are critical to rapidly develop policy interventions that mitigate COVID-19, to anticipate impacts on health care resources, and to strategize how best to impose and lift public health guidelines. Existing COVID-19 models produce radically different predictions, thus creating confusion and mistrust over their use. Therefore, there is an urgent need to compare between the wide-range of existing COVID-19 models and their predictions. The goal of this project is to find a consensus among various model predictions and to make the different model assumptions and uncertainty transparent. An interactive web-based dashboard will serve as an open and accessible tool to inform the public, fellow researchers, and decision-makers where existing models agree and disagree on predictions. The predictions that are agreed upon by existing models can then be used with greater confidence and trust as the basis for effective decision-making to save lives and resources. Apart from the practical importance for the implementation of effective pandemic control measures and public health strategies, other broader impacts are professional development opportunities for early career researchers and training opportunities for a post-doctoral scholar. To find a consensus among various model predictions, this project will develop a novel ensemble prediction approach that (1) aligns different COVID-19 simulation models and (2) uses a time series clustering technique to unify model predictions. In the model alignment stage, a range of open-source and publicly available COVID-19 simulation models will be selected, aligned based on their parameters, and run as needed. A broad set of predictions will be obtained from the model results, where each prediction represents a possible world with a corresponding number of new cases, fatalities, and other quantities of interest. The time series clustering stage will project possible worlds into a feature space and apply clustering algorithms to find similar possible worlds. For each cluster, a representative possible world will be selected, enriched with measures of uncertainty, and visualized using the dashboard. This project advances the theoretical knowledge base for model alignment approaches and representative uncertain clustering for simulation model predictions. The unification of model predictions into a scientific consensus can be used to inform decision-makers better so that they can develop life-saving interventions.This RAPID award is made by the Ecology and Evolution of Infectious Diseases Program in the Division of Environmental Biology, using funds from 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.