OmniOta validation in four pages

Shigella counts as Escherichia: it is polyphyletic within E. coli and its ANI exceeds the species threshold; the name is retained for clinical, not taxonomic, relevance (ISO 13136:2012). It affects 36 of the 78 reads in the Ecoli dataset — without the merge the figure would be 39.7% instead of 94.6%, and that is why it is declared. There are exactly two cases in the whole catalogue, they derive from the complexes the product already shows the user, and they do not touch species names.

The dataset staggered is HM-783D from BEI Resources: twenty certified species between 0.03% and 41.25%, with congeners that have to be separated —S. aureus de S. epidermidis, three Streptococcus— and members three orders below the dominant one. 96.86% at genus, 185 of 191 reads, an upper bound. Brown published 93% on this same material. It does not enter the mean because GAIA published no figure for it.

Ten species with composition certified in the manufacturer's datasheet — eight bacteria at 12% genomic DNA and two yeasts at 2% — sequenced by Nicholls, Quick, Tang and Loman (2019) on GridION and PromethION. 50,000 reads measured per platform.

The composition is invariant between GridION and PromethION: the per-genus differences run from 0.05 to 0.43 points and the global divergence is statistically indistinguishable from sampling. What does not repeat are the low-abundance artefacts —Macrococcus, which is not a member, appears with 23 reads on one platform and 5 on the other. The composition repeats, the traces do not, and that settles how much weight to give a taxon held up by twenty reads. Accuracy is an upper bound because this is a mixture; the reads come from the run's prefix, which over-represents the flow cell's first hours, and it is stated as such.

A 2026 product evaluated on 2019 chemistry is a legitimate objection, and it is not answered with arguments but by measuring. Identity is measured by aligning against the standard's genomes; it is not taken from the specification sheet.

On that chemistry the engine classifies 89.5% of the reads at genus with 98,53 % accuracy and 57.5% at species with 97,26 %, finding 7 of 7 members at both ranks. Against the 2019 material, reads resolved to species go from 31.1% to 57.5% at the same accuracy: nearly twice the resolution without losing precision.

Between Brown 2017 and Zymo D6300 everything changes at once: year, chemistry, sequencer, community, laboratory and identity. With two points differing in six variables it is not justified to attribute a collapse to one of them. The experiment takes the same file from Zymo and injects error at increasing rates with a fixed seed: everything else is identical by construction. Substitutions only, no indels — which are more destructive for an exact k-mer — so the scenario is conservative in the competitor's favour.

The most degraded level lands at 74.23%, the median identity of Brown 2017: no 2015 dataset was needed; degrading the 2019 one is enough. All eight organisms are recovered at every level, including the most severe, where the competitor does not return a single line. It is the result that matters for a real food sample: DNA fragmented by processing, a fatty matrix, difficult extraction. Genus accuracy rises as the material degrades (99.07 → 99.40%) and that is not a merit but selection bias: the easy reads are what remain, and classified reads fall from 53.7% to 36.6%. On modern chemistry sylph works well: it detects 7 of 7 members on all three basecallers, including fast.

Against GAIA only its published figures are available: closed service, proprietary database. With sylph the reference can indeed be equalised, and it has been, genome by genome, on certified D6300. Both tools find the complete community and each is stronger on a different axis.

Per read OmniOta is highly precise, and that is the figure governing an abundance table and everything computed on top of it. Per named taxon sylph decides better: it models genome coverage with k-mer statistics and requires real presence before naming, whereas we name any taxon that receives an assignment. The criterion under development is a minimum evidence to name, not merely to assign: the same Wilson doctrine the product already applies to per-taxon confidence, carried into the decision to publish.