Summary
A bioRxiv preprint reports that simulated mixtures of E. coli strains made sequence-type and serotype prediction less reliable than detection of individual resistance and virulence genes. Increasing the dominant strain’s abundance improved strain-level predictions, while longer reads alone did not resolve the challenge.
A bioRxiv preprint from researchers at Uppsala University reports that genomic tools recovered individual antimicrobial-resistance and virulence genes more consistently than they identified the strains present in simulated mixtures of Escherichia coli. The benchmark found that the mixture’s strain composition affected multilocus sequence typing (MLST) and serotype prediction, even when some genetic markers could be detected.
What the benchmark found
The researchers created simulated, single-species E. coli metagenomes from genomes with annotations determined by PCR. They compared read-based and assembly-based tools for detecting seven antimicrobial resistance genes (ARGs) and three virulence genes (VGs), as well as for MLST and serotype prediction. MLST assigns a sequence type using a defined set of genetic markers; serotyping classifies bacteria by features associated with their surface antigens.
Across the evaluated tools, ARG recovery ranged from 84.3% to 90.0%. The read-based tool SRST2 had the highest reported virulence-gene recovery, at 100.0%. By comparison, complete serotype recovery was uncommon across methods. In a multi-strain dataset with a log-normal distribution of strain abundances, none of the evaluated MLST tools correctly identified the dominant sequence type, although some MLST alleles were recovered.
The balance of strains mattered. When the simulated mixture was made less complex by increasing the dominant strain’s abundance, both MLST and serotype prediction improved. Analyses using longer reads indicated that read length alone was not enough to overcome the difficulty of resolving a mixed-strain population.
Why gene detection and strain typing differ
A gene-detection result asks whether a particular sequence, such as a resistance determinant, is present. Strain typing asks which combination of markers belongs together in a particular bacterial strain. In a mixture, genetic material from several strains is present at once, making that assignment more difficult. As a result, reliable recovery of individual genes does not necessarily mean that the strains carrying them have been identified accurately.
That distinction matters for genomic surveillance: a profile of detected resistance genes can be more dependable than a strain-level account of who carries them. The authors present their benchmark as a framework for investigating how strain abundance and population structure affect reliable strain-resolved analysis.
The work is a preprint based on simulated, single-species E. coli metagenomes. Its reported performance therefore describes the benchmark conditions, rather than establishing how the same tools perform across all real-world samples or bacterial communities.