Benchmarking strain-level profiling of Escherichia coli in short-read gut metagenomes

StrainR2 achieved the highest F1 score (0.995) for Escherichia coli strain-level profiling, distinguishing all strains in a complex 10-strain spike-in within a 304-strain reference database.

• Why it matters: Choosing among reference-based short-read metagenomic tools is unclear, yet accurate strain composition matters for understanding Escherichia coli colonisation diversity and distinguishing infection from carriage.
• What they did: Six tools (PanTax, PathoScope 2.0, StrainGE, Strainify, StrainR2, StrainScan) were benchmarked on real and simulated metagenomes with increasing complexity and varying reference database composition, assessing strain detection and relative abundance.
• The result: When ground-truth strains were removed from the reference database, PathoScope 2.0 and StrainGE performed strongest for phylogroup abundance accuracy and frequent correct strain identity assignment, with median runtime 0.11–68.45 h and peak memory 2.08–351.53 GiB.