Embracing eDNA and machine learning for taxonomy-free microorganisms biomonitoring to assess the river ecological status

Abstract

Currently implemented biomonitoring and bioassessment are an indispensable part of river management worldwide. Although DNA-based biomonitoring is a promising tool for biodiversity detection, a substantial effort still exists about how DNA-derived microorganisms’ datasets can be used for routine bioassessment. Here, we analyzed microorganisms (i.e., bacteria and microbial eukaryotes) collected from the Dongjiang River, to develop a Metabarcoding-eDNA Index (MEI) using the taxonomy-free strategy based on supervised machine learning, and to identify the comparability of the new MEI index with two traditional water pollution indices (i.e., trophic state index, TSI, and water quality index, WQI) for assessing river ecological status. First, our data showed that a rich diversity of microorganisms in rivers was detected by the eDNA technology, yet 40–90% of OTUs were unknown across taxa because of the deficiency of reference databases and the inability to annotate. Second, combining quantile regression and supervised machine learning, ca. 90% of unknown OTUs were reliably classified into different ecological groups corresponding to the TSI and WQI. Third, the new MEI index had significant and consistent correlations with the TSI (Rbacteria2 = 0.74; Reukaryotes2 = 0.69) and WQI (Rbacteria2 = 0.64; Reukaryotes2 = 0.53). Overall, by incorporating the eDNA-derived microorganisms’ datasets with supervised machine learning, this study further validates the reliability and robustness of taxonomy-free biomonitoring, supporting the routine application of microorganisms for river ecological status assessment.