利用人类移动性推断传染病数据中的位置并校正偏差
Inferring Locations and Correcting Bias in Infectious Disease Data Using Human Mobility
Owen Tucker, Lily Wang, Ruochen Kong, Haley Stone, Jing Du, Matthew Scotch, Flora D. Salim, Taylor Anderson 等 10 位作者
AI 速读 基于摘要
作者提出将病原体基因组数据与人类移动性数据结合,用移动感知的信念传播方法推断未观测传播节点的位置,并在模拟与真实基因组监测数据上报告了交叉熵的下降。
cs.SI
查看原文摘要
Understanding and reconstructing infectious disease transmission remains challenging due to sparsely sampled and biased epidemiological case data: only a small fraction of infectious disease cases are reported to health authorities, and an even smaller fraction are sent to labs for genomic sequencing. In this work, we study the problem of inferring unobserved transmission events by integrating pathogen genomic data with human mobility information. Intuitively, even if we do not have case data for a certain region, such as a census block group, we can use massive-scale individual mobility data to understand where people may be exposed and infer where unobserved cases may be located. Assuming only a subset of cases are reported and sequenced, we reconstruct a phylogenetic tree from observed cases and transform it into a partial transmission tree using the outbreaker2 tool. This tree contains inferred transmissions based on the genetic distance of the observed cases. As these inferred nodes have unknown locations, we propose a mobility-aware belief propagation approach to estimate their locations using human mobility data. Experiments on simulated and real-world genomic surveillance data with biased sampling show that incorporating mobility data improves the reconstruction of inferred transmission trees, reducing cross-entropy relative to the observed distribution by 39.4% and 16.1%, respectively. Results on real-world COVID-19 sequence data show the promise of combining genomic and mobility data for more reliable location inference, with potential implications for outbreak analysis and public health decision-making.