Benchmarking 16S rRNA gene amplicon analysis in high-diversity microbial communities reveals fundamental trade-offs in clustering and denoising

A study has examined the performance of four popular 16S rRNA gene amplicon analysis pipelines—VSEARCH cluster_size, UNOISE in VSEARCH, Swarm, and DADA2—in characterising highly diverse environmental microbial communities.

The researchers benchmarked these pipelines using simulated microbial communities with established compositions, gauging how well they performed as species richness, abundance unevenness, sequencing depth, and minimum abundance threshold fluctuated. Initially, variations between pipelines were minimal at low species richness. However, as species richness rose, the differences became more pronounced.

Notably, high average cluster purity did not always translate to precise species-level recovery. Some species were fragmented across multiple clusters, while others were only partially recovered. UNOISE emerged as a potential contender when prioritising high species representation and cluster purity, although it tended to split species into numerous clusters and generated a substantial number of unclustered reads.

DADA2 demonstrated strong relative abundance profile reconstruction with minimal species splitting but represented fewer species and exhibited lower cluster purity at higher richness levels.

Swarm offered a balanced middle ground between species representation and splitting, while cluster_size may be more suitable when limiting species splitting is a priority, despite weaker relative abundance profile reconstruction. Increasing the minimum abundance threshold curtailed species splitting but also reduced the number of clusters and perfect clusters and, at elevated thresholds, diminished alignment with ground-truth relative abundance profiles. These effects were more pronounced at lower sequencing depths.

Field validation demonstrated sequence loss, including the disappearance of sequences that were frequently detected across replicate samples. This phenomenon was contingent on the minimum abundance threshold.

The study concludes that the performance of these pipelines varies depending on dataset characteristics, evaluation metrics, and parameter choices. Consequently, pipeline selection should be dictated by the specific dataset traits and analytical goals rather than by a single performance metric. Furthermore, minimum abundance thresholds need to be evaluated in relation to sequencing depth rather than being arbitrarily fixed.

These insights underscore the necessity of reevaluating current analytical practices and the importance of openly reporting bioinformatic settings as 16S rRNA amplicon sequencing becomes more commonplace in diverse environmental microbial community studies.