RAPID: A fast and scalable method to improve epidemiological models for COVID-19

当前的新冠肺炎大流行正在促使科学界改进目前用于了解传染病传播的流行病学模型。目前的模型将个体划分为易感、暴露但非传染性、无症状但传染性、症状和传染性、康复和死亡等区间。为了简化传染病的数学建模，流行的假设是同一隔间中的个体行为相同。给这些所谓的隔间模型增加复杂性的一种方法是以更细粒度的方式对个体进行分类，从而增加更多的隔间。这会导致向模型中添加更多参数。估计这些参数需要更多的数据，而更多的数据增加了估计参数的计算成本。此外，正如这场大流行所显示的那样，我们获得数据的方式多种多样。在每个国家和直辖市内，正在执行不同的抽样战略。随着更多的数据可用，该项目降低了建立和更新复杂的地区性流行病学模型的计算成本。通过这样做，该项目提高了科学界对病毒传播做出更准确预测的能力，并通报了关于缓解战略的地方政策决策的有效性。研究人员采用了最近在分子动力学模拟分子建模中取得成功的一类模型拟合-最大熵偏置方法。这些方法用与模型参数无关的最小偏置项来代替模型参数优化。这使得模型优化的运行时间复杂度与数据量呈线性关系，与未知参数个数无关。这些活动将使复杂模型能够快速优化，这些模型还考虑了空间分辨率和采样偏差。优化过程的改进成本将允许频繁地更新室模型，而不需要在每次观察到新数据时进行完全参数优化。调查人员将通过公布代码、数据和发现，与迅速合并的新冠肺炎研究社区中的其他人密切合作。这一奖项反映了美国国家科学基金会的法定使命，并已通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The current COVID-19 pandemic is prompting the scientific community to improve the epidemiological models currently employed to understand the spread of infectious diseases. Current models divide a population of individuals into compartments, such as, susceptible, exposed but non-infectious, asymptomatic but infectious, symptomatic and infectious, recovered, and deceased. To simplify the mathematical modeling of infectious diseases, the prevailing assumption is that individuals in the same compartment behave identically. One way to add sophistication to these so-called compartmental models is by categorizing individuals in a more fine-grained manner, thus adding more compartments. This results in more parameters added to a model. Estimating these parameters requires more data, and more data increases the computational cost of estimating the parameters. In addition, as this pandemic is showing, our access to data is varied. Within each country and municipality, different sampling strategies are being pursued. This project lowers the computational cost of setting up and updating complex, compartmental epidemiological models as more data becomes available. By doing so, the project improves the ability of the scientific community to make more accurate predictions on the spread of the virus and inform on the effectiveness of local policy decisions on mitigation strategies. The investigators adopt a category of model fitting that has seen recent success in molecular dynamics simulations in molecular modeling — maximum entropy biasing methods. These methods replace model parameter optimization with a minimal biasing term that is independent of the model parameters. This makes the runtime complexity of model optimization linear with the amount of data and independent of the unknown number of parameters. The activities will enable rapid optimization of complex models that additionally consider spatial resolution and sampling biasing. The improved cost of the optimization process will permit frequent updates of compartmental models without the need for full parameter optimization each time new data is observed. The investigators will collaborate closely with others in the rapidly coalescing COVID-19 research community by releasing code, data, and findings.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.