Comparing transmission reconstruction models with Mycobacterium tuberculosis whole genome sequence data

Abstract

Pathogen genomic epidemiology is now routinely used worldwide to interrogate infectious disease dynamics. Multiple computational tools that reconstruct transmission networks by coupling genomic data with epidemiological modelling have been developed. The resulting inferences are often used to inform outbreak investigations, yet to date, the performance of these transmission reconstruction tools has not been compared specifically for tuberculosis, a disease process with complex epidemiology that includes variable latency periods and within-host heterogeneity. Here, we carried out a systematic comparison of seven publicly available transmission reconstruction tools, evaluating their accuracy in predicting transmission events in both simulated and real-world Mycobacterium tuberculosis outbreaks. No tool was able to fully resolve transmission networks, though both the single-tree and multi-tree input implementations of TransPhylo identified the most epidemiologically supported transmission events and the fewest false positive links. We observed a high degree of variability in the transmission networks inferred by each approach. Our findings may inform the choice of tools in future tuberculosis transmission analyses and underscore the need for caution when interpreting transmission networks produced using probabilistic approaches.

“Outbreak reconstruction with a slowly evolving multi-host pathogen: a comparative study of three existing methods onMycobacterium bovisoutbreaks.”

“…With the low mutation rate characteristic of M. bovis, we estimated poor accuracies (median accuracy lower than 9% for all three methods). Sobkowiak et al compared these outbreak reconstruction methods on real M. tuberculosis data, which is also a slow-evolving pathogen, and estimated the positive predictive value (PPV), meaning the number of epidemiologically linked case-contact pairs that were correctly identified (preprint, [54]). Contrary to the accuracy indicator we estimated, the links between cases were not directed, we thus expected this study to estimate a higher number of correctly reconstructed cases.…”

Using genetic data to identify transmission risk factors: Statistical assessment and application to tuberculosis transmission

“…Our findings are consistent with a recent benchmarking study of computational methods for reconstructing TB transmission, which included TransPhylo and five other methods. [29] That study used simulations to show low sensitivity for correctly identifying transmission events under realistic low-TB burden scenarios for all methods tested. Another recent study used simulations to evaluate the use of genomic data and GLM for identifying risk factors for TB transmission, defined as genetic closeness (<2 SNPs).…”

Integrating Transmission Dynamics and Pathogen Evolution Through a Bayesian Approach

“…This gap underscores a need in transmission inference: a model that can account for transmission bottleneck, complete transmission history, and within-host diversity, as well as provide uncertainty intervals for its estimates. See also recent work by Duault, Durand, andCanini 2022 andSobkowiak et al 2022 for a more detailed review and comparison of the available methods.…”