Study on Robustness of Denoising and Clusterization Algorithms for the Computation of Microbiome Data

One crucial factor for dissimilarity of microbiome studies is the choice between denoising or clusterization algorithms, respectively. Moreover, the robustness, or stability of these algorithms with respect to the number of sequences computed, and its effect on the calculated ecological metrics of the microbiome studied, are currently unknown. In this study, mock communities were used for the investigation of robustness of several denoising and clusterization algorithms. The effect that denoising error models and clusterization similarity thresholds, and the effect of rare biosphere removal from ecological analyses, were also assessed. The results showed robustness of α-diversity metrics, while β-diversity was clearly affected differentially for each of the algorithms tested. Taxonomic accuracy showed differences with respect to algorithms, number of sequences computed, and taxonomic closeness of the community studied. Removal of rare biosphere affected values of α-diversity and β-diversity metrics and taxonomic accuracy differently for algorithms and datasets tested. Clusterization and denoising algorithms showed different degrees of stability, which were influenced greatly by the dataset, and possibly the taxonomic similarity of microorganisms in the community studied. The results obtained are important for the future configuration of universal standards for microbiome exploration as well as for current methodological applications in this field.