Advances in DNA sequencing technology have stimulated the rapid uptake of protocols-such as eDNA analysis and metabarcoding-that infer the species composition of environmental samples from DNA sequences. DNA barcode reference libraries play a critical role in the interpretation of sequences gathered through such protocols, but many of these libraries lack a taxonomic consensus, include redundant records, do not support end-user analytical pipelines, and are not permanently archived. Furthermore, because DNA sequencers are outpacing Moore's Law and reference libraries are growing, the computational power required to assign sequences to source taxa is rapidly increasing. This paper introduces an algorithmic approach to construct DNA barcode reference libraries that addresses these issues. Hosted online, 'BOLDistilled' libraries are comprehensive but compact, because the algorithm distills genetic variation into a minimal set of records. We provide a BOLDistilled library for the barcode region of the cytochrome c oxidase 1 gene (COI) based on data in the Barcode of Life Data System (BOLD). It contains 1.7 M records versus the 15.7 M in the complete library, a compression that reduced the time required for sequence analysis of metabarcoded samples by ≥ 98% with no reduction in the accuracy of taxonomic placements. BOLDistilled libraries will be updated regularly, with current and previous versions available at https://boldsystems.org/data/boldistilled. By providing access to persistent, comprehensive, and high-quality reference data, these libraries strengthen the capacity of DNA-based identification systems to advance biodiversity science.
Prosser et al. (Sun,) studied this question.