Investigators reported that differences between segment trees can arise from tree reconstruction error alone. The team assembled nearly 12,000 avian H5Nx genomes and focused on how topology discordance translates into inferred reassortment. Short, clear finding: simulations that contained no reassortment still produced substantial numbers of subtree prune and regraft (SPR) events when trees were reconstructed from sampled HA sequences. This indicates a tangible false-positive contribution to reassortment counts derived from standard tree-edit metrics.
The authors curated 11,795 complete avian H5Nx genome sequences and used segment 4, encoding hemagglutinin (HA), as the anchor for their approach. They sampled subsets of HA sequences chosen to yield different expected levels of phylogenetic reconstruction error. From each sampled HA set the group reconstructed a reference tree; these reference trees then served as the scaffolds for downstream sequence simulation. The source did not report additional procedural details here beyond the sampling rationale and use of HA as the reference segment.
Using the HA reference trees, the study team simulated molecular evolution along those topologies to generate sequences matching the lengths and evolutionary rates of the other genomic segments. Because simulations were performed without any reassortment, any discordance observed between the simulated segment trees and the HA reference must be attributable to reconstruction error. Data indicated that simulations from random HA samples produced about 32% as many SPRs as were inferred from real data, while simulations from the most genetically divergent HA subset reduced that proportion to 4%. These comparisons quantify specificity loss driven by reconstruction uncertainty and by sampling composition.
To evaluate sensitivity, the investigators modified simulated trees by introducing known SPR rearrangements and then assessed the rate at which those engineered events were recovered. The authors reported a false-negative rate of 8.5% for detection of introduced SPRs. They also observed that the false-negative rate increased with the number of SPRs per tree, indicating reduced power to detect reassortment as event density rises. The source did not supply exact counts of introduced SPRs per tree in this summary; measurements were reported in aggregate as a false-negative fraction.
The group compared entanglement — an untangling optimization metric for tanglegrams — and SPR distances between the HA reference tree and trees for other segments in both real and simulated datasets. They computed distance-to-SPR values using Espalier and included distance-to-siblings and entanglement results in the repository. Data indicated that a nontrivial fraction of tree discordance observed in empirical IAV segment comparisons can be reproduced by simulations that lack reassortment, establishing reconstruction error as a measurable contributor to apparent reassortment signals.
Beyond counts of SPRs, the study characterized inferred rearrangements according to the size of the moved subtree and the distance traversed within the tree. The authors analyzed whether certain classes of SPRs — for example, small subtrees moved short distances versus large subtrees moved long distances — were more likely to be genuine reassortment events versus artifacts of reconstruction error. The source indicates these characterizations were performed but does not provide granular numeric breakdowns within the article summary; the associated Zenodo repository contains the entanglement results and SPR-distance computations used for this assessment.
The authors emphasized that phylogenetic uncertainty and sampling design materially affect reassortment inference from tree discordance. They showed reassortment can be substantially overestimated when reconstruction error is not accounted for, and they quantified both false-positive and false-negative contributions under their simulation framework. The study made its sequence accession lists, computed entanglement values, distance-to-SPR outputs, and analysis scripts available in a Zenodo repository (DOI provided in the source). The source did not report specific methodological parameters such as the exact simulation settings per segment within this summary text, nor did it present granular per-class SPR counts here. Remaining unanswered questions include how these error rates generalize across different IAV subtypes, host species beyond avian, and alternative tree-reconstruction pipelines; those specific extensions were not reported in the article summary.
Data availability and reproducibility are strengths: the authors provided the raw sequences and all analysis materials in the linked repository. The work therefore allows other groups to replicate the simulation framework and to test how alternative sampling or reconstruction strategies alter false-positive and false-negative rates for reassortment detection in influenza A virus datasets.