Supervised machine learning outperforms taxonomy‐based environmental DNA metabarcoding applied to biomonitoring

Abstract

Biodiversity monitoring is the standard for environmental impact assessment of anthropogenic activities. Several recent studies showed that high-throughput amplicon sequencing of environmental DNA (eDNA metabarcoding) could overcome many limitations of the traditional morphotaxonomy-based bioassessment. Recently, we demonstrated that supervised machine learning (SML) can be used to predict accurate biotic indices values from eDNA metabarcoding data, regardless of the taxonomic affiliation of the sequences. However, it is unknown to which extent the accuracy of such models depends on taxonomic resolution of molecular markers or how SML compares with metabarcoding approaches targeting well-established bioindicator species. In this study, we address these issues by training predictive models upon five different ribosomal bacterial and eukaryotic markers and measuring their performance to assess the environmental impact of marine aquaculture on independent data sets. Our results show that all tested markers are yielding accurate predictive models and that they all outperform the assessment relying solely on taxonomically assigned sequences. Remarkably, we did not find any significant difference in the performance of the models built using universal eukaryotic or prokaryotic markers. Using any molecular marker with a taxonomic range broad enough to comprise different potential bioindicator taxa, SML approach can overcome the limits of taxonomy-based eDNA bioassessment.

Studying Ecosystems With DNA Metabarcoding: Lessons From Biomonitoring of Aquatic Macroinvertebrates

“…For this reason, interest in taxonomy-free approaches is increasing among those studying poorly-known assemblages whose morphological identification is challenging (e.g., meiofauna or diatoms; Vasselon et al, 2017). Clearly defining the unit and universe of observation (i.e., taxonomic breadth and resolution) is fundamental to comparing such characteristics (Cordier et al, 2018;Pawlowski et al, 2018), but doing so could also improve compatibility between biogeographically separated programs (Turak et al, 2017;Bailet et al, 2019). Nonetheless, to tie DNA-based monitoring to historic surveys, and to assign ancillary information such as traits, it is still a requirement to assign taxonomic names to identified sequences (e.g., Compson et al, 2018).…”

Quantifying the information content of lake microbiomes using a machine learning-based framework

“…The results and methods presented in this study represent an important contribution to the discussion around the use of microorganisms in lake ecosystem monitoring schemes. Firstly, they indicate that the physico-chemical status of a lake cannot fully be predicted by its microbiome (see figures 1, 2B), even if the microbiome can be used for biomonitoring [33]. Nevertheless, up to around 60% of the variation in certain parameters can be predicted by the lakes microbial community composition, which is comparable to results from soil ecosystems [92].…”

“…Our main contribution is a machine learning-based framework for the quantification of the information shared between the microbiome and a total of 25 physico-chemical and positional (i.e., GPS coordinates and altitude) parameters of an ecosystem. It builds upon a wealth of studies that elucidate the role of the microbiome in ecology using machine learning [19,[31][32][33][34][35][36][37][38][39]. In our framework, a model learns a projection of the microbial prevalence space to a single dimension for each of the parameters, which makes it able to handle the extremely high dimensionality of amplicon-based microbiome datasets.…”

Responses of foraminifera communities to aquaculture‐derived organic enrichment as revealed by environmental DNA metabarcoding

“…Although eDNA metabarcoding has been shown to be a powerful and efficient alternative tool to monitor benthic impacts of fish farms, its application for routine regulatory benthic impact monitoring is still hotly debated (Cordier et al., , ; Forster et al., ; Keeley, Wood, & Pochon, ; Pawlowski et al., , ; Pochon et al., ; Stoeck, Kochems, Forster, Lejzerowicz, & Pawlowski, ; Stoeck, Frühe et al., ). In the countries where sediments are assessed using biological indices that require benthic faunal characterization, eDNA metabarcoding can replace traditional morpho‐taxonomy to calculate biological indices once there are enough reference sequences for benthic species in public databases and the ecological value of the targeted group of species is known (Pawlowski et al., , ).…”

“…Populating reference databases is slow and labour intensive, and does not get much attention. An alternative taxonomy‐free approach proposed to overcome this limitation is supervised machine learning, which can predict the ecological quality of sediments based on eDNA metabarcoding data of unclassified sediments and a training dataset (Cordier et al., , ). For countries where routine benthic monitoring relies heavily on geochemistry data (Cranford, Brager, & Wong, ), eDNA metabarcoding could significantly enhance existing approaches by providing a rapid and accurate means of generating biological data from sediments.…”