TLife-LSTM: Forecasting Future COVID-19 Progression with Topological Signatures of Atmospheric Conditions
TLife-LSTM: Forecasting Future COVID-19 Progression with Topological Signatures of Atmospheric Conditions
复制标题
DOI：
10.1007/978-3-030-75762-5_17
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
I. Segovia-Dominguez;Zhiwei Zhen;R. Wagh;Huikyo Lee;Y. Gel
中科院分区：
文献类型：
--
作者：
I. Segovia-Dominguez;Zhiwei Zhen;R. Wagh;Huikyo Lee;Y. Gel
了解大气条件对 SARS-CoV2 的影响对于模拟 COVID-19 动态至关重要，并有助于了解未来在世界范围内的传播。此外，COVID-19 预期临床严重程度的地理分布可能与既往呼吸道疾病史以及湿度、温度和空气质量的变化密切相关。在这种情况下，我们假设通过跟踪一段时间内大气条件的拓扑特征，我们可以提供可能与 COVID-19 动态相关的大气变化的可量化结构分布。因此，我们在图的时间序列上应用持久同源机制来提取拓扑特征并跟踪相对湿度和温度的地理变化。我们开发了一个名为拓扑寿命 LSTM (TLife-LSTM) 的综合机器学习框架，并测试其预测 SARS-CoV2 病例动态的预测能力。我们使用美国华盛顿州和加利福尼亚州记录的确诊病例数和住院率来验证我们的框架。我们的结果证明了 TLife-LSTM 在预测 COVID-19 动态和对其复杂时空传播动态建模方面的预测潜力。
Understanding the impact of atmospheric conditions on SARS-CoV2 is critical to model COVID-19 dynamics and sheds a light on the future spread around the world. Furthermore, geographic distributions of expected clinical severity of COVID-19 may be closely linked to prior history of respiratory diseases and changes in humidity, temperature, and air quality. In this context, we postulate that by tracking topological features of atmospheric conditions over time, we can provide a quantifiable structural distribution of atmospheric changes that are likely to be related to COVID-19 dynamics. As such, we apply the machinery of persistence homology on time series of graphs to extract topological signatures and to follow geographical changes in relative humidity and temperature. We develop an integrative machine learning framework named Topological Lifespan LSTM (TLife-LSTM) and test its predictive capabilities on forecasting the dynamics of SARS-CoV2 cases. We validate our framework using the number of confirmed cases and hospitalization rates recorded in the states of Washington and California in the USA. Our results demonstrate the predictive potential of TLife-LSTM in forecasting the dynamics of COVID-19 and modeling its complex spatio-temporal spread dynamics.