Paula Magbor, Western University

Paula Magbor (Western University) and Art Poon (Western University)

Discovering & Evolving

Virus transmission rates are often studied by generating networks linking genetically similar infections. By convention, component clustering (CC) is used to extract subsets of infections from the network. However, CC can obscure the risk heterogeneity within large components, e.g., two cliques connected by a single edge are one component. We investigated the application of community detection (CD) methods to resolve this issue. We implemented multi-deme compartmental models in TiPS, and simulated HIV-like sequences with Pyvolve. Additionally, we retrieved and aligned 12,556 HIV-1 pol sequences collected in China from Genbank, filtering for non-overlapping sequences. The TN93 genetic distances were computed for each alignment, and the resulting pairwise matrix was used to construct networks under different thresholds (0.005-0.045 substitutions/site). We extracted clusters from these networks using seven methods in CDlib, and computed the adjusted mutual information (AMI) to measure concordance between clusters and putative risk factors. We obtained higher AMI (0.271-0.297) for CD methods at higher thresholds (TN93=0.035) than CC (AMI=0.165, TN93=0.025) on simulations, demonstrating a benefit to partitioning large components. For actual HIV-1 data, AMI was also higher for CD (0.137-0.145) than CC (0.042) at higher TN93 thresholds. CD also showed higher concordance with sampling locations by province. These findings demonstrate the utility of CD methods in detecting latent transmission risk structures within genetic clusters. Improving our understanding of viral transmission dynamics is essential for guiding the prioritization of treatment and prevention strategies.