A method for detecting horizontal gene transfer events of antibiotic resistance genes using phylogenetic trees

Abstract

An algorithm, named the HGT-score algorithm, to computationally assess horizontal gene transfer of antibiotic resistance genes spread was developed using a methodology based on phylogenetic trees and host taxonomy labels. Applying the HGT-score algorithm on validation data showed that the algorithm detects horizontal gene transfer events. This work serves as a good starting point for further investigation and improvements.

The results are two-fold, consisting of the developed methods themselves, and the results that the methods generated when applied on real data. Antibiotic resistance gene (ARG)-based phylogenetic trees were cross-sectioned at different lengths from the root, named cutoffs. A cutoff generates flat clusters of leaves, encoding ARG similarity between hosts within each cluster. Distantly related host bacteria may end up in the same cluster if their genomes contain identical or similar antibiotic resistance genes. HGT-scores were given to clusters depending on the taxonomic similarity of member hosts, with higher dissimilarity resulting in a higher HGT-score.

The method was validated on a labelled dataset of ARGs, where the labels sorted ARGs into HGT-prone genes and non-HGT-prone genes. The HGT-prone data scored higher than non-HGT labelled data, particularly in specific cutoff regions. It is therefore plausible that the method scores HGT events higher, and that the cutoff is an important factor to account for in drawing conclusions as to when the events occurred.

Lastly, there are several directions to explore in order to improve the method. The HGT-score method is dependent on three key parameters: cutoff, cluster-level statistic, and p. These parameters should all be investigated and calibrated for different use case scenarios. A variant of the method should be investigated, using well-conserved genes rather than taxonomy as the base of gene tree-host species dissimilarity. This opens up the possibility of detecting HGT events across more closely related bacteria. Another interesting direction is fine-tuning cutoffs. Finer cutoffs could provide better resolution in critical parts of the phylogenetic gene trees where many bifurcations occur.