Files
Abstract
As genomic data become increasingly feasible nowadays, transmission events are often identified by a phylogenetic tree estimated from pathogen genomes. However, due to incomplete lineage sorting, the estimated phylogenetic tree may be incongruent with the transmission network for the diseases with a long incubation period or the pathogens with high mutation rates. Moreover, many methods for identification of transmission events treat symptom onset time as the actual infection time, which may mislead the statistical inference of the transmission network. Here we present a Bayesian model to reconstruct outbreaks of densely sampled tuberculosis using genomic data and temporal time information while considering within-host diversity and latent periods. The simulation study indicates that the Bayesian model yields an averaged 85% ~ 98% prediction accuracy in the estimation of transmission networks under several scenarios. As the number of single nucleotide polymorphisms (SNP) increases to infinity, the Bayesian model can consistently reconstruct the true transmission network. Our method assumes a complete sampling of infected individuals. Nevertheless, an outbreak is rarely perfectly sampled in reality. Reconstruction of a real-world tuberculosis outbreak in Uganda displayed little uncertainty owing to the large variation presents in SNP data. This implies the transmission events estimated from genomics may not be the direct infections between individuals. Therefore, we develop a hypothesis testing to identify direct transmissions within the estimated network by comparing the expected and the observed numbers of SNPs. These results suggest that the Bayesian model accounting for latent periods and within-host diversity is useful for understanding transmissions of pathogens among human populations.