Work

Reference augmentation in Bangladeshi fish DNA barcoding: cross-programme label agreement and discordant assignments

Year: 2026
Type: preprint
Abstract: DNA barcode reference augmentation can increase agreement with species labels while introducing discordant assignments. We compared 98 conservatively grouped public Bangladeshi marine/coastal finfish cytochrome c oxidase subunit I records across 33 published labels and two collection-programme blocks. Reciprocal comparisons used a frozen international library of 124 reference groups, with 63 or 35 groups added from the other Bangladeshi programme. Queries, libraries and benchmark rules were locked before primary predictions, with historical development exposure disclosed. Offline BLASTn and alignment-distance matching used a common 450-base core and uncalibrated acceptance rules. Both methods increased accepted agreement from 24/35 to 30/35 for University of Dhaka queries and from 46/63 to 53/63 for the linked Sher-e-Bangla Agricultural University programme. The latter net gain combined 11 abstentions becoming agreements and four agreements becoming disagreements, all four involving one added snapper reference. A post-hoc sensitivity expanded the international library to 536 eligible cached groups within the same label scope. The augmentation gain shrank to 4/35 for Dhaka queries (25/35 to 29/35) and remained 7/63 for the other programme (46/63 to 53/63). Larger-library arms had no accepted disagreements, but species ties produced abstentions, including all five reciprocal snapper conflict records. The multinomial naive-Bayes comparator had higher observed agreement, accompanied by accepted disagreements and uncalibrated confidence. A restricted post-hoc count-matched comparison retained 30 and 59 queries: local versus international additions agreed for 26/30 versus 20/30 and 49/59 versus 42/59, respectively. These fixed-library observations show why agreement, disagreement and abstention should be reported together. They establish neither independently verified biological accuracy nor a Bangladesh-specific causal benefit. The study repository and supporting materials are available on request.
Language: English
Subfield: Ecology
Open Access status: green
OpenAlex ID: W7236803380
Cites:
Cited by:
Related to: 10