Predicting the epidemic threshold of the susceptible-infected-recovered model

TL;DR

This paper compares theoretical methods for predicting epidemic thresholds in networks, finding that the dynamical message passing method generally provides more accurate predictions, but performance varies with network structure.

Contribution

It systematically analyzes and compares the accuracy of MFL, QMF, and DMP methods across different network types, revealing conditions where each method performs best.

Findings

DMP method yields more accurate epidemic thresholds on uncorrelated networks.

DMP predictions are closer to real-world thresholds due to topology and correlations.

MFL method performs better in networks with certain structural features.

Abstract

Researchers have developed several theoretical methods for predicting epidemic thresholds, including the mean-field like (MFL) method, the quenched mean-field (QMF) method, and the dynamical message passing (DMP) method. When these methods are applied to predict epidemic threshold they often produce differing results and their relative levels of accuracy are still unknown. We systematically analyze these two issues---relationships among differing results and levels of accuracy---by studying the susceptible-infected-recovered (SIR) model on uncorrelated configuration networks and a group of 56 real-world networks. In uncorrelated configuration networks the MFL and DMP methods yield identical predictions that are larger and more accurate than the prediction generated by the QMF method. When compared to the 56 real-world networks, the epidemic threshold obtained by the DMP method is closer to the actual epidemic threshold because it incorporates full network topology information and some dynamical correlations. We find that in some scenarios---such as networks with positive degree-degree correlations, with an eigenvector localized on the high $k$-core nodes, or with a high level of clustering---the epidemic threshold predicted by the MFL method, which uses the degree distribution as the only input parameter, performs better than the other two methods. We also find that the performances of the three predictions are irregular versus modularity.