Reads from a standard whose species were absent from the reference database were added to a sample of known composition. For each method, the resulting abundance profile was compared with the corresponding profile without added reads. A profile shift of zero indicates that the added reads did not alter the estimated composition.
profile shift = Σ |share with added reads − share without| over the catalogue species
Half this distance represents the minimum fraction of abundance redistributed between profiles (0.17 ≈ 8.5%). % added: added reads as a proportion of the complete mixture (at 50%, added and original reads are equal in number); column headers report the added-read counts. Samples: all 14,978 pooled Microbial reads; 4,992 reads drawn from the pooled Oral runs (seqtk 1.3, seed 42). Bold: smallest shift in each column.
| Method | 10% added 1,664 reads | 25% added 4,993 reads | 50% added 14,978 reads |
|---|---|---|---|
| Kraken2 | 0.002 | 0.003 | 0.008 |
| Bracken | 0.001 | 0.004 | 0.009 |
| BASEN-nf | 0.000 | 0.000 | 0.000 |
| BASEN | 0.000 | 0.000 | 0.000 |
| Method | 10% added 555 reads | 25% added 1,664 reads | 50% added 4,992 reads |
|---|---|---|---|
| Kraken2 | 0.005 | 0.013 | 0.043 |
| Bracken | 0.005 | 0.013 | 0.042 |
| BASEN-nf | 0.000 | 0.001 | 0.002 |
| BASEN | 0.000 | 0.000 | 0.000 |
| Method | 10% added 555 reads | 25% added 1,664 reads | 50% added 4,992 reads |
|---|---|---|---|
| Kraken2 | 0.024 | 0.061 | 0.168 |
| Bracken | 0.024 | 0.060 | 0.167 |
| BASEN-nf | 0.003 | 0.007 | 0.022 |
| BASEN | 0.000 | 0.000 | 0.003 |
The table reports the number of added reads assigned to a catalogue species by Kraken2 and the subset retained by BASEN after read filtering. Kraken2, Bracken and BASEN-nf use every classified read (Bracken excludes only species with fewer than 10 reads). The final column reports the proportion of classified reads from the original sample removed by the filter. Classification proportions were similar at 10% and 25% (3.3–3.6% unrelated, 15–16% related).
| Mixture 50% added: equal numbers of added and original reads | Added reads classified by Kraken2 all misassigned | Classified added reads retained by BASEN passed the read filter | Sample's own reads removed by the filter |
|---|---|---|---|
| Microbial sample (14,978 reads), Oral reads added (unrelated) | 487 (3.3%) | 0 | 3.3% |
| Oral sample (4,992 reads), Microbial reads added (unrelated) | 182 (3.6%) | 0 | 1.8% |
| Oral sample (4,992 reads), Gut reads added (related) | 750 (15.0%) | 18 | 1.8% |
The table reports the five most frequent taxonomic assignments among added reads classified by Kraken2 before read filtering. Because their source species are absent from the database, these classifications are considered false-positive species-level assignments. For related (Gut) reads, genus-level assignments were largely concordant with their origins: alignment to the Gut reference genomes using Minimap2 indicated that 437 of the 441 reads assigned to the Oral V. parvula originated from the Gut V. rogosae. The shared species name requires qualification: 206 of the 209 reads assigned to the Oral F. nucleatum originated from the Gut F. nucleatum, which has the same species name but shares only 91.0% ANI with the Oral genome, below the 95% species threshold, and was therefore treated as an absent taxon. The filter retained none of the classified unrelated reads and 18 of the 750 classified related reads (section 2).
| Assigned to | Reads (%) | |
|---|---|---|
| Microbial sample (14,978 reads), Oral reads added (unrelated) (487 classified) | ||
| Enterococcus faecalis | 97 (20%) | |
| Pseudomonas aeruginosa | 61 (13%) | |
| Escherichia coli | 54 (11%) | |
| Listeria monocytogenes | 49 (10%) | |
| Bacillus subtilis | 45 (9%) | |
| Oral sample (4,992 reads), Microbial reads added (unrelated) (182 classified) | ||
| Streptococcus mutans | 44 (24%) | |
| Streptococcus salivarius | 40 (22%) | |
| Haemophilus parainfluenzae | 39 (21%) | |
| Streptococcus parasanguinis | 18 (10%) | |
| Streptococcus | 11 (6%) | |
| Oral sample (4,992 reads), Gut reads added (related) (750 classified) | ||
| Veillonella parvula | 441 (59%) | |
| Fusobacterium nucleatum | 209 (28%) | |
| Haemophilus parainfluenzae | 39 (5%) | |
| Prevotella nigrescens | 20 (3%) | |
| Neisseria subflava | 13 (2%) | |
K-mer support was calculated for each Kraken2-classified read and summarized as the median across reads:k-mer support = k-mers matching the assigned taxon / all k-mers of the read
Kraken2 can classify an entire read from a small number of matching k-mers, whereas BASEN credits a species only with the k-mers supporting that assignment. For a read of 1,000 bp (966 k-mers), the median support observed for original-sample reads corresponds to about 840–890 supporting k-mers, compared with about 3–67 for misassigned added reads. These reads contribute fully to Kraken2 and Bracken read counts but contribute little to BASEN-nf, explaining the limited profile shifts in section 1 without read filtering.
| Mixture | Sample's own reads % of k-mers supporting the assignment | Added reads (all misassigned) % of k-mers supporting the assignment |
|---|---|---|
| Microbial sample (14,978 reads), Oral reads added (unrelated) | 87.4% | 0.3% |
| Oral sample (4,992 reads), Microbial reads added (unrelated) | 92.6% | 0.9% |
| Oral sample (4,992 reads), Gut reads added (related) | 92.6% | 6.9% |