Answers are grounded in this paper and cite their sources.
Multi-barcode analysis improves species identification accuracy in gastropods, indicating enhanced biodiversity assessment capabilities.
DNA barcode‐based species identification has become pivotal in biodiversity research. Current barcoding‐based deep learning methods have advantages in species identification but exhibit significant limitations. By using COI barcoding alone, previous studies were unable to fully capture species identification features. Prior methods showed that high classification accuracy at the species level was high but reduced accuracy for higher taxonomic levels. We constructed a multi‐type barcoding dataset, including COI, 16S, H3, 18S, ITS1 and ITS2, for taxa of the two most species‐rich subclasses within Gastropoda, Caenogastropoda and Heterobranchia. Multi‐type barcoding was represented by RNA secondary structure embedding, and COI was additionally featured by DNA barcode embedding. To address uneven sample coverage among barcoding types, we designed a three‐stage curriculum learning framework by training COI data, freezing parameters, fusing low‐coverage barcoding data and then full tuning to obtain the final model, SnailBaLLsp. By employing a hierarchical attention network that propagates higher taxonomy‐level classifications as features to lower levels, SnailBaLLsp improved the classification accuracy across all taxonomy levels. SnailBaLLsp achieved 84.99% species‐level and 92.93% genus‐level accuracy for Gastropoda. Multi‐barcode analysis outperformed single COI barcode at all taxonomy levels. The three‐stage strategy boosted the unseen species prediction accuracy by 3.79%, while further fusion of the additional DNA barcoding embedding of COI improved it by 8.55%, surpassing state‐of‐the‐art baselines across all levels. SnailBaLLsp effectively predicted label‐unknown case data, with failure predictions in species‐level linked to high intraspecific sequence divergence. It also exhibits strong cross‐taxon transferability, as shown by its high performance after fine‐tuning on multi‐barcoding data of Bivalvia. By successfully utilizing RNA secondary structure while also incorporating DNA barcoding features, we processed multi‐type barcoding data with incomplete coverage while providing multi‐taxonomy‐level species predictions from subclass to species, offering an advanced deep learning‐based solution in the molecular barcoding aspect for integrated taxonomy.
No takes yet. Share an insight, caveat, or question.
Ye et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: