Epidemic Source Detection in Contact Tracing Networks: Epidemic Centrality in Graphs and Message-Passing Algorithms

An Overview of Healthcare Data Analytics With Applications to the COVID-19 Pandemic

“…Designing computational techniques that exploit statistical features is thus key to algorithmic speedup without incurring significant information loss or degraded solution quality. We give an overview of Network Centrality as Statistical Inference [103,104,105,106] and Graph Neural Networks as useful frameworks to design scalable algorithms for digital contact tracing and other computational epidemiology optimization problems.…”

“…In other words, the forward engineering approach enables the reuse of existing algorithms or a performance comparison between different graph algorithms. It also provides a message-passing (i.e., belief propagation) algorithmic perspective to improve existing network centrality-based algorithms [103,104,105,106].…”

“…We describe briefly a Contact Tracing Algorithm 4 in [104] that uses a weighted distance centrality measures where the weights are computed for node regression using the aforementioned GNN approach to solve (2). As an illustration, by reconstructing the contact tracing network data of SARS-CoV2003 (a virus very similar to the COVID-19 coronavirus) in Taiwan [104], this algorithm can correctly identify the first place, Taipei Municipal Heping Hospital (now Taipei City Hospital Heping Branch), of an infection cluster in April 2003 in Taiwan as compared to a breadthfirst search heuristic in [103], which chooses the red vertex modeling a confirmed index case (not the first case) who had been to the Taipei Municipal Heping Hospital as shown in Figure 3(a). In addition, the network centrality approach can enable visualization tools (e.g., dot distribution map) to visualize the likelihood of infection source that may be of value to public healthcare policymakers.…”

“…Input: Infection networks {G 1 , G 2 , ..., G N } harvested by forward contact tracing Output: A ranking of the outbreak source probability of each node for backward contact tracing 1: Calculate the structure features (e.g., degree ratio and infected proportion ratio) for each node in the network. 2: Calculate the labels (i.e., approximate solution to (2)) for each node according to a network centrality algorithm (e.g., rumor centrality in [103,105] or statistical distance centrality algorithm in [104]). 3: Train the regression model by GraphSage [111] to output a ranking of the probability for each node to be the outbreak source in each network {G 1 , G 2 , ..., G N }.…”

“…In systems biology, [123] applied PSO to select biological model and estimate parameters in the model, and [124] gave a review of metaheuristics for estimating parameters. Likewise, they are also increasingly used in disciplines [104] and in [103] respectively. (b) Photo courtesy of the World Health Organization Infodemic risk management system in [8] to assess the spread of COVID-19 related misinformation.…”