Sourmash accurately profiles strains from long-read metagenomes preserving gene content

Sourmash strain profiling of long-read mock metagenomes preserves gene content and estimates abundances accurately, with gene recovery F1 scores from 0.942 to 0.998 across five communities.

• Why it matters: Strain-level metagenomics is hard because strains can share >99% ANI yet differ in gene content, so nucleotide similarity alone may mislead functional inference and strain identification.
• What they did: The study benchmarked sourmash using k-mer sketches with FracMinHash against reference databases on five mock communities of increasing complexity with known composition, evaluating strain identification, gene content recovery, and abundance estimation.
• The result: When exact strains were not recovered, sourmash returned exactly or closely related strains with highly conserved gene content, and abundance estimates matched expected compositions with Pearson correlations of 0.823–1 at strain level and 0.978–1 at species level.

AI summary of the abstract · may contain errors · not medical advice · read the paper.