Debar: A sequence‐by‐sequence denoiser for COI‐5P DNA barcode data

Abstract

DNA barcoding and metabarcoding are now widely used to advance species discovery and biodiversity assessments. High-throughput sequencing (HTS) has expanded the volume and scope of these analyses, but elevated error rates introduce noise into sequence records that can inflate estimates of biodiversity. Denoising -the separation of biological signal from instrument (technical) noise-of barcode and metabarcode data currently employs abundance-based methods which do not capitalize on the highly conserved structure of the cytochrome c oxidase subunit I (COI) region employed as the animal barcode. This manuscript introduces debar, an R package that utilizes a profile hidden Markov model to denoise indel errors in COI sequences introduced by instrument error. In silico studies demonstrated that debar recognized 95% of artificially introduced indels in COI sequences. When applied to real-world data, debar reduced indel errors in circular consensus sequences obtained with the Sequel platform by 75%, and those generated on the Ion Torrent S5 by 94%. The false correction rate was less than 0.1%, indicating that debar is receptive to the majority of true COI variation in the animal kingdom. In conclusion, the debar package improves DNA barcode and metabarcode workflows by aiding the generation of more accurate sequences aiding the characterization of species diversity.

VLF: An R package for the analysis of very low frequency variants in DNA sequences

“…Similar computational and statistical tools, in the form of MATLAB packages, R packages, Python packages and methodological pipelines, used to assess anomalies in DNA (meta)barcodes, have been released. Examples include divisive hierarchical clustering: DADA ( Rosen et al 2012 ) and DADA2 ( Callahan et al 2016 ); artificial neural networks: ( Ma et al 2018 ); Profile Hidden Markov Models: coil ( Nugent et al 2020 ), debar ( Nugent et al 2021 and Porter and Hajibabaei 2021 ); distribution sample quantiles: MACER ( Young et al 2021 ); and Shannon entropy: SequenceBouncer ( Dunn 2021 ), A2G2 ( Hleap et al 2020 ), DnoisE ( Antich et al 2022 and Turon et al 2020 ). These methods and programmes are beginning to see widespread use within the biodiversity and regulatory science communities.…”

Machine learning in molecular ecology

“…Machine learning approaches also allow prediction of patterns of biodiversity at large geographical scales by facilitating the combination of genomic, ecological and geographical data in novel ways (Barrow et al, 2020). Finally, machine learning can improve our estimates of biodiversity by allowing efficient error correction of barcoding data sets (Nugent et al, 2021). Again, the diversity of ap-…”

“…However, errors can result in inflated estimates of diversity when not corrected. To address this issue,Nugent et al (2021) introduce debar, an approach for denoising COI-5P DNA barcode data using machine learning. Debar uses a Profile Hidden Markov model (PHMM) to detect indel errors in COI barcoding data.…”