Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Global Prediction of COVID-19 Variant Emergence Using Dynamics-Informed Graph Neural Networks

Majd Al Aawar, Srikar Mutnuri, Mansooreh Montazerin, Ajitesh Srivastava

TL;DR

The paper tackles predicting when a new COVID-19 variant will reach a target prevalence in a region before local emergence. It develops a dynamics-informed Graph Neural Network (GNN) that embeds derived variant growth rates and time-series prevalence ratios into a mobility-based country graph, and contrasts it with a PINN-inspired baseline. The FA-GCN architecture, which fuses temporal dynamics with graph structure, achieves superior delay prediction performance over multiple baselines, while ablations underscore the value of adjacency-based connectivity and dynamics features. A retrospective benchmark pipeline over 87 countries and 36 variants is introduced to standardize evaluation and spur further research. The work provides a practical, scalable framework for global variant emergence forecasting and public health preparedness.

Abstract

During the COVID-19 pandemic, a major driver of new surges has been the emergence of new variants. When a new variant emerges in one or more countries, other nations monitor its spread in preparation for its potential arrival. The impact of the new variant and the timings of epidemic peaks in a country highly depend on when the variant arrives. The current methods for predicting the spread of new variants rely on statistical modeling, however, these methods work only when the new variant has already arrived in the region of interest and has a significant prevalence. Can we predict when a variant existing elsewhere will arrive in a given region? To address this question, we propose a variant-dynamics-informed Graph Neural Network (GNN) approach. First, we derive the dynamics of variant prevalence across pairs of regions (countries) that apply to a large class of epidemic models. The dynamics motivate the introduction of certain features in the GNN. We demonstrate that our proposed dynamics-informed GNN outperforms all the baselines, including the currently pervasive framework of Physics-Informed Neural Networks (PINNs). To advance research in this area, we introduce a benchmarking tool to assess a user-defined model's prediction performance across 87 countries and 36 variants.

Global Prediction of COVID-19 Variant Emergence Using Dynamics-Informed Graph Neural Networks

TL;DR

The paper tackles predicting when a new COVID-19 variant will reach a target prevalence in a region before local emergence. It develops a dynamics-informed Graph Neural Network (GNN) that embeds derived variant growth rates and time-series prevalence ratios into a mobility-based country graph, and contrasts it with a PINN-inspired baseline. The FA-GCN architecture, which fuses temporal dynamics with graph structure, achieves superior delay prediction performance over multiple baselines, while ablations underscore the value of adjacency-based connectivity and dynamics features. A retrospective benchmark pipeline over 87 countries and 36 variants is introduced to standardize evaluation and spur further research. The work provides a practical, scalable framework for global variant emergence forecasting and public health preparedness.

Abstract

During the COVID-19 pandemic, a major driver of new surges has been the emergence of new variants. When a new variant emerges in one or more countries, other nations monitor its spread in preparation for its potential arrival. The impact of the new variant and the timings of epidemic peaks in a country highly depend on when the variant arrives. The current methods for predicting the spread of new variants rely on statistical modeling, however, these methods work only when the new variant has already arrived in the region of interest and has a significant prevalence. Can we predict when a variant existing elsewhere will arrive in a given region? To address this question, we propose a variant-dynamics-informed Graph Neural Network (GNN) approach. First, we derive the dynamics of variant prevalence across pairs of regions (countries) that apply to a large class of epidemic models. The dynamics motivate the introduction of certain features in the GNN. We demonstrate that our proposed dynamics-informed GNN outperforms all the baselines, including the currently pervasive framework of Physics-Informed Neural Networks (PINNs). To advance research in this area, we introduce a benchmarking tool to assess a user-defined model's prediction performance across 87 countries and 36 variants.
Paper Structure (31 sections, 18 equations, 7 figures, 3 tables, 2 algorithms)

This paper contains 31 sections, 18 equations, 7 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work and Background
  3. Variants and Variant Dynamics
  4. Graph Neural Networks and Epidemics
  5. Methodology
  6. Variant Dynamics
  7. Variant Growth Rates
  8. Dynamics-based Model
  9. Dynamics-Informed GNN
  10. Graph Construction
  11. Feature Augmented Graph Convolutional Network (FA-GCN)
  12. Physics Informed GCN (PI-GCN)
  13. Training Procedure
  14. Experiments
  15. Dataset
  16. ...and 16 more sections

Figures (7)

  • Figure 1: Plots showing the prevalence of a few different variants of COVID-19 over time in the UK and Sweden. The beginning of the spread of new variants can differ across countries by several weeks. This can be seen when examining "20I.Alpha.V1" and "21J.Delta" variants in the provided plots, showing their appearance in the UK several weeks before reaching Sweden.
  • Figure 2: Sample semi-log plots of variant proportions from data in the CoVariants dataset.
  • Figure 3: Graph creation process on sample subgraph $\mathcal{G}$: (i) first we construct our nodes and edges based on country adjacency (ii) Next, we account for temporal variations in the relations between countries, i.e. the edges,to get a $\mathcal{G}_t$ (iii) Finally, we find the variant specific features to get our sample graph $\mathcal{G}_{t,i}$
  • Figure 4: FA-GCN Architecture. The prevalence ratio for each time step is encoded into a 2D latent feature vector by the GRU. Then, the feature vectors are flattened and the growth rate is concatenated to them. Our batch size of $N$ = # of countries.
  • Figure 5: Benchmarking Pipeline
  • ...and 2 more figures