Predicting sediment ecological state from metagenomes shows equal performance for taxonomic and functional features

The use of environmental microbial DNA to monitor the ecological state in seafloor sediments has many advantages and efforts are being made to find reliable biomarkers from DNA-based taxonomic profiles. However, the taxonomic composition of microbial communities can vary over time and space, while their functional characteristics typically remain consistent. Furthermore, functionality may better capture the breadth of biological complexity. Therefore, we here tested whether functional attributes of microbial communities serve as more reliable indicators of environmental quality than their taxonomic composition. To test this, we analyzed a set of Metagenome-Assembled-Genomes (MAGs) from 41 different coastal locations in Norway and Iceland, characterized by environmental impact gradients resulting from salmon aquaculture. Functional and taxonomic features extracted from these MAGs were then used to predict the ecological state of the corresponding sample sites using several supervised machine learning models and stratified feature selection. Our findings indicate that both taxonomic and functional features demonstrated comparable effectiveness in predicting environmental quality. This outcome has direct relevance for eDNA-based regulatory compliance monitoring. However, the functional insights derived from the most significant functional features identified by machine learning models remain essential for deepening our understanding of the ecological processes underpinning practical biomonitoring tools.