Linked machine learning models improve species classification of fungi when using error-prone long-reads on extended metabarcodes

The increased usage of long read sequencing has not been matched with publicly available databases suited for error-prone long reads when applied to metabarcodes. We address this gap and present a new method for classifying species using linked machine learning models. We demonstrate its capability for classifying species with high accuracy, show the benefit of this approach over current alignment and k-mer methods for classifying very closely related species, and suggest a confidence score cutoff of 0.85 to improve the potential for accurately identifying a target species from a mixed sample of unknown composition. Finally, we suggest future applicable use of the approach in medicine, agriculture, and biosecurity.