Sampling and summarizing transmission trees with multi-strain infections

Abstract

Motivation The combination of genomic and epidemiological data holds the potential to enable accurate pathogen transmission history inference. However, the inference of outbreak transmission histories remains challenging due to various factors such as within-host pathogen diversity and multi-strain infections. Current computational methods ignore within-host diversity and/or multi-strain infections, often failing to accurately infer the transmission history. Thus, there is a need for efficient computational methods for transmission tree inference that accommodate the complexities of real data. Results We formulate the direct transmission inference (DTI) problem for inferring transmission trees that support multi-strain infections given a timed phylogeny and additional epidemiological data. We establish hardness for the decision and counting version of the DTI problem. We introduce Transmission Tree Uniform Sampler (TiTUS), a method that uses SATISFIABILITY to almost uniformly sample from the space of transmission trees. We introduce criteria that prioritize parsimonious transmission trees that we subsequently summarize using a novel consensus tree approach. We demonstrate TiTUS’s ability to accurately reconstruct transmission trees on simulated data as well as a documented HIV transmission chain. Availability and implementation https://github.com/elkebir-group/TiTUS. Supplementary information Supplementary data are available at Bioinformatics online.

Methods Combining Genomic and Epidemiological Data in the Reconstruction of Transmission Trees: A Systematic Review

“…Moreover, intrinsic characteristics (predominant species, number of animals, production period) are only considered in two methods, belonging to the NPF and SimPF, respectively: Aldrin 2011 [28] and BORIS (Bayesian Outbreak Reconstruction Inference and Simulation) [29] (Table 1). Similarly, only two other methods included contact data in their transmission model: one in the NPF, outbreaker2 [30], and one in the SeqPF, TiTUS [31] (Table 1). Ten out of the twenty-two methods (Table 1) were implemented in packages, 7 available as R packages (however, since their implementation, two have been removed from the CRAN repository; for details see Table S6), one code on github (Transmission Tree Uniform Sampler, TiTUS), and the remaining two on BEAST [13] (beastlier) or BEAST2 [14] (Structured COalescent Transmission Tree Inference, SCOTTI).…”

“…The non-phylogenetic family estimated transmission probability from calculated pairwise genetic distances. However, two families used phylogenetic trees to reconstruct transmission trees, either by inferring the host of each node or branch in the phylogenetic tree [2,17,18,31,40,41], considering within-host phylogenetic trees as part of a hierarchical tree [5,26], or by using the phylogenetic tree as a source of information [39]. In the sequential phylogenetic family, phylogenetic trees were reconstructed prior to the implementation of the method and thus called for an additional choice, the phylogenetic tree reconstruction method.…”

“…The possibility of transmitting genetically diverse strains was overlooked in most methods due to a strong assumption, that is, a transmission bottleneck size of one transmitted sequence [37]. This assumption was relaxed in three methods, , for whom transmission bottleneck size varied [37] 2020), who disregarded transmission bottlenecks completely, allowing the transmission of multiple strains [31,41] and even multiple infections in SCOTTI [41].…”

“…Similarly, Sashittal and El-Kabir (2020) aimed to label the internal nodes in a phylogenetic tree reconstructed from HIV sequences [31] (node annotation similar to Figure 4A). However, in this method, a weak transmission bottleneck was considered.…”

“…While the transmission process was not explicitly modeled (Table 3), the labelling had to satisfy a number of constraints derived from the known transmission windows (i.e., from exposure time to removal time) and contact information (Table 1). The transmission tree reconstruction was treated as a logical problem and a parsimonious consensus tree was then selected from uniformly sampled transmission trees that satisfied the temporal and contact constraints [31].…”