16S reanalysis · Ng et al. 2019 humanised mice

Reproducibility: QIIME 2 against nf-core/ampliseq

The same checksummed per-sample reads were processed by nf-core/ampliseq 3.1.0 with DADA2 settings matched to QIIME 2, and the frozen analysis was rerun. The decision rule was tagged before the second workflow ran.

−0.1454QIIME 2 contrast
−0.1453nf-core/ampliseq contrast
0.0001difference between workflows (prespecified tolerance 0.02)
788shared ASVs, carrying ≥ 99.99% of reads

Figure 4

Cage values of the frozen primary measure in the two workflows.

Scatter of the five cage values of the frozen primary measure, QIIME 2 on one axis and nf-core/ampliseq on the other; all points lie on or very close to the line of equality.

Each point is one cage: the cage mean of the Bray–Curtis distance between each mouse’s days 8–12 samples and its own day-0 sample, computed from QIIME 2 (horizontal) and nf-core/ampliseq (vertical). STD = standard diet; PD = low-MAC diet (the paper’s MD arm). The dashed line marks equal values. The diet contrast differed by 0.0001 between workflows (−0.1454 against −0.1453), against a prespecified tolerance of 0.02, and the cage pattern was the same.

Comparison

CheckResult
Read retentionFinal read count per sample, nf-core/ampliseq ÷ QIIME 2: median 1.000 (range 0.983–1.001)
ASVs788 shared, carrying 99.99% of QIIME 2 reads and 100.00% of ampliseq reads (rounded); the rest are rare
Same sample, two workflowsBray–Curtis median 0.0000, maximum 0.017 (two truncation settings within QIIME 2: median 0.024)
Frozen contrast−0.1453 against −0.1454: same direction; difference 0.0001 against a tolerance of 0.02

Agreement shows reproducibility under matched processing, not robustness to analysis choices; robustness was assessed separately with sensitivity analyses.