# Prediction of genetic relatedness of Escherichia coli using neighbor typing: a tool for rapid outbreak detection

## TL;DR

This study shows that a rapid neighbor typing method with long-read sequencing can accurately predict the genetic relatedness of E. coli, helping detect outbreaks quickly.

## Contribution

The study introduces a rapid neighbor typing method paired with long-read sequencing for accurate and fast genetic relatedness prediction in E. coli.

## Key findings

- Strong correlations were found between neighbor typing and reference methods (Spearman’s rho = 0.75–0.95).
- Using a lineage score-informed approach improved correlations to Spearman’s rho = 0.93–0.94.
- Predicted genetic trees were comparable to reference methods with high cluster and tree similarity metrics.

## Abstract

Identifying the genetic relatedness of resistant bacterial pathogens in healthcare settings can help identify undetected transmission events and outbreaks. However, current methods are time- and resource-intensive. We evaluated a rapid neighbor typing method paired with long-read sequencing for assessment of genetic relatedness. Utilizing a data set of primary clinical samples and published isolate data from two outbreaks of Escherichia coli, we applied genomic neighbor typing of long-read sequence data to rapidly estimate genetic relatedness. We assessed the correlation between neighbor typing predicted genetic distance and pairwise genetic distance from short-read draft whole genomes for all sample pairs. Predicted genetic trees using neighbor typing were compared to reference genetic trees generated using mash distances and maximum-likelihood (ML) methods to assess the extent of agreement, along with metrics of cluster similarity (cluster comparability and Baker’s gamma index [BGI]) and tree topology similarity (generalized Robinson-Foulds [GRF] metric). For all three data sets, we found strong correlations between the reference methods and predicted genetic distances (Spearman’s rho = 0.75–0.95, P < 0.001), which improved when using a lineage score-informed approach (Spearman’s rho = 0.93–0.94, P < 0.001). Predicted genetic trees and clusters from neighbor typing were comparable to those generated using either mashtree or an ML method, with a range of cluster comparability of 85.8–99.5%, BGIs of 0.8–0.95, and GRF values of 0.34–0.8. Pairing the neighbor typing method with long-read sequencing can enable accurate predictions of the relatedness of E. coli samples and isolates, and could potentially be used as a rapid outbreak surveillance tool.