HiFiTaxa: a reproducible analysis pipeline for full length 16S rRNA gene and ITS amplicon sequencing from PacBio reads

HiFiTaxa is a new, reproducible Nextflow pipeline designed for analyzing PacBio reads containing full-length 16S rRNA gene and ITS amplicons. This tool addresses the challenge posed by taxonomic accuracy in long-read amplicon sequencing, which hinges on the choice of reference, classifier, and parameters employed.

The creators of HiFiTaxa have constructed versioned references using the latest GTDB and UNITE releases and have incorporated two different classifiers: Bayesian lowest common ancestor (BLCA) and naive Bayes (NB), alongside the read-level profilers Emu for 16S rRNA gene and EMITS for ITS. To demonstrate its effectiveness, HiFiTaxa underwent rigorous benchmarking against mock communities and holdouts with 10% of GTDB or UNITE database sequences withheld.

When applied to a 16S rRNA gene mock community with 20 species, the BLCA classifier identified all 20 species but reported two false positives. In contrast, the NB classifier detected 17 species without any errors. The Emu profiler managed to identify all 20 species but also reported eight false positives. In the case of a GTDB holdout, BLCA achieved a remarkable 80.0% accuracy in species-level identification, while the NB classifier achieved 48.1%.

However, the NB classifier was found to be more conservative when dealing with novel taxa, while EMITS demonstrated the highest detection rate for mock species.

In summary, HiFiTaxa provides a single, reproducible workflow that leverages current GTDB and UNITE releases, granting users the flexibility to select between different classifier algorithms based on their specific requirements.