Rank-biased overlap – an alternative method to measure community similarity from compositional data

Measuring the similarity, or β-diversity, of communities is a fundamental part of many ecological analyses. However, most traditional similarity metrics fail to account for compositionality, making them unsuitable for the taxa count tables generated through molecular approaches such as metabarcoding. Recently, distance metrics that first transform the data to abundance ratios have been proposed, such as the robust Aitchison distance, but the performance of these metrics is disputed, with few alternative similarity metrics suggested. Here, we propose the use of a rank-based similarity metric, rank-biased overlap (RBO), as a non-parametric alternative to measure community similarity from compositional datasets. RBO can handle incomplete, non-conjoint, and top-weighted ranked lists; all of which are properties that molecular taxa tables commonly exhibit. To characterise the properties of RBO for ecological analyses, we benchmarked RBO alongside several commonly used similarity metrics (Jaccard, Bray-Curtis, and Sorensen distance) and robust Aitchison distance, against an existing suite of criteria using simulated data. In most criteria, RBO performed at least as well as traditional metrics, and almost always outperformed robust Aitchison distance. While other metrics were able to better reflect shifts in species abundance distributions, RBO showed greater scale-invariance than most others, including robust Aitchison distance, making it ideal for molecular datasets where library sizes are unavoidably variable. Overall, RBO has a number of desirable properties that make it ideal for measuring similarity from compositional molecular community datasets and should be considered a robust non-parametric similarity metric suitable for metabarcoding data.