Cooperating Graph Neural Networks With Deep Reinforcement Learning for Vaccine Prioritization

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Current vaccine distribution models overlook population mobility, leading to suboptimal micro-level allocation.
• Limited vaccine supply necessitates efficient prioritization strategies to minimize pandemic impact.

Purpose of the Study:

• To develop an optimal vaccine allocation strategy incorporating mobility heterogeneity.
• To reduce the overall burden of a pandemic under limited vaccine supply.

Main Methods:

• Proposed a Trans-vaccine-SEIR model to integrate mobility patterns into disease propagation.
• Developed a deep reinforcement learning framework utilizing graph neural networks for optimal allocation.
• Incorporated mobility network structures and disease features for enhanced strategy.

Main Results:

• The proposed framework reduced infections and deaths by 7%-10% compared to baseline strategies.
• Demonstrated robustness across diverse mobility patterns.
• Identified transit usage restrictions as more effective than cross-zone mobility restrictions for high-risk zones.

Conclusions:

• The novel deep reinforcement learning framework provides an effective vaccine allocation strategy considering mobility.
• Findings offer critical insights for resource-limited public health interventions and pandemic response.