Epidemic control under incomplete observations through mechanistic state reconstruction and multi-agent learning
PLOS Computational Biology · 16.11.2026 · 22 (11) · e1014955
DOI: 10.1371/journal.pcbi.1014955 ↗Kurz gefasst
Fachartikel in der Zeitschrift PLOS Computational Biology, online erschienen am 16.11.2026. Verfasst von Shang Wang und 5 weiteren Autor:innen. Beteiligte Einrichtung: Guangdong Provincial Center for Disease Control and Prevention, Guangzhou, China. Die Arbeit nennt 53 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Shenzhen Key Basic Research Project, Guangdong Provincial Center for Disease Control and Prevention.
Abstract (Originalsprache)
Epidemic-control policies are usually optimized from reported infection states, but these reports are often incomplete because testing capacity, reporting behavior, delays, and asymptomatic infections vary across regions and time. Such observation gaps can distort the state used by the controller and lead to suboptimal testing and isolation decisions. We develop a three-stage computational epidemiology framework that links observation modeling, state reconstruction, and policy optimization. First, we extend a metapopulation SEIQR model to represent fully observable, steady partially observable, and non-steady partially observable surveillance regimes. Second, we introduce ODE-DynNet, a hybrid reconstruction model that combines mobility-coupled epidemic dynamics with graph-temporal learning to infer hidden infection states from distorted observations. Third, we formulate coordinated testing and isolation as a multi-agent reinforcement learning problem with a flow-based responsibility-aware reward that attributes transmission costs across connected regions. Using community-level mobility data from Shenzhen, China, we show that ODE-DynNet reduces infection cost by 88% (from 4,930 to 590) relative to the non-steady partially observable, decoupled policy that uses the raw proxy without reconstruction. The same pipeline remains effective across transmission intensities and reporting-delay settings, indicating that explicit observation modeling and state reconstruction are useful preprocessing steps for learned epidemic-control policies under incomplete surveillance.
Autor:innen und Institutionen
- Guangdong Provincial Center for Disease Control and Prevention, Guangzhou, China.
- Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China.
- University of Chinese Academy of Sciences, Beijing, China.
- The Hong Kong Polytechnic University, Hong Kong, China.
- Faculty of Computer Science and Control Engineering, Shenzhen University of Advanced Technology, Shenzhen, China.
ORCID-Profile: Shang Wang, Ling Yin
Förderung
- National Natural Science Foundation of China
- Shenzhen Key Basic Research Project
- Guangdong Provincial Center for Disease Control and Prevention
Zitieren
APA
Wang, S., Luo, Y., Cui, Y., Zhang, M., Kang, M., & Yin, L. (2026). Epidemic control under incomplete observations through mechanistic state reconstruction and multi-agent learning. PLOS Computational Biology, 22(11), e1014955. https://doi.org/10.1371/journal.pcbi.1014955
Vancouver
Wang S, Luo Y, Cui Y, Zhang M, Kang M, Yin L. Epidemic control under incomplete observations through mechanistic state reconstruction and multi-agent learning. PLOS Computational Biology. 2026;22(11):e1014955. doi:10.1371/journal.pcbi.1014955
BibTeX
@article{Wang2026,
author = {Wang, Shang and Luo, Yuxiao and Cui, Yunduan and Zhang, Meng and Kang, Min and Yin, Ling},
title = {Epidemic control under incomplete observations through mechanistic state reconstruction and multi-agent learning},
journal = {PLOS Computational Biology},
year = {2026},
volume = {22},
number = {11},
pages = {e1014955},
doi = {10.1371/journal.pcbi.1014955}
}
EUVIMED