Fifth step criteria were blast match length ≥ 0.69, number mismatches ≤ 4, number gaps ≤ 1 and percent identity ≥ 99%.
Sixth step criteria included blast match length ≥ 0.625, number mismatches ≤ 8, number of gaps ≤ 1, and percent identity ≥ 98%.
Fourth step criteria were blast match length ratio ≥ 0.725, number mismatches ≤ 5, number of gaps = 0, and percent identity ≥ 99%.
Seventh step criteria included blast match length ratio ≥ 0.575, number mismatches ≤ 13, number gaps ≤ 2 and percent identity ≥ 97%.
The eighth step criteria were blast match length ≥ 0.52, number mismatches ≤ 12, number of gaps ≤ 2 and percent identity ≥ 97%.
The third step criteria were blast match length ratio ≥ 0.87, number of mismatches ≤ 4, number of gaps = 0, and percent identity ≥ 99%.
We propose that better classification results can be obtained by tuning the taxonomic classifier to each BLAST HSP length, reference gene, and taxonomic rank.
In order to balance recall and precision while identifying cross-hybridizing probes for which the target gene requires a substitute SSP on the H&M Array, the optimal set of BLAST predictions (length of the alignments and the number of mismatches), are determined using the ideal F-Score while systematically varying each parameter independently (Additional file 1 - Supp. Figure S2).
Reads having BLAST match lengths <90% with the representative sequence in the three databases were considered as chimeras and removed.
On the other hand, mean Blast alignment lengths using the 454 unigenes are approximately 2× shorter than for the Sanger unigene set (data not shown).
