Prokaryote community metabarcoding data from sediments of coastal and estuarine locations of the Bay of Biscay

Routine monitoring of benthic biodiversity is critical for managing and understanding the anthropogenic impacts on marine, transitional and freshwater ecosystems. However, traditional reliance on morphological identification generally makes it cost-prohibitive to increase the ambition of monitoring programs. Metabarcoding of environmental DNA has a clear potential to overcome many of these issues, with prokaryotes and other micoorganisms showing particular promise as indicator organism. However, due to the limited knowledge regarding their ecological roles and responses to different types pressure, de novo approaches are necessary. Here, we use two such approaches for the prediction of multiple impacts present in estuaries and coastal areas of the Bay of Biscay and evaluate our results using cross-validation. The first (Random Forests) is a machine learning method while the second (Threshold Indicator Taxa Analysis and quantile regression splines) is based on classical bio-indication. We also analyse and compare the most important taxa identified by both approaches. Our results show that both methods overlap considerably in the indicator taxa utilised, and perform well in spite of the complexity of the studied ecosystem, providing predictive models with strong correlation to reference values and fair to good agreement with established ecological status. The ability to predict several specific types of pressure including time-integrative impact is especially appealing. The cross-validated models and biotic indices developed can be directly applied to new data from estuaries of the same geographical region, although more work is needed to evaluate and improve them for use in new regions or habitats.