Fig. 5: Comparison of the relative importance of viral and bacterial communities for the nutrient cycling during HTC.

a Percentage increases in the MSE (mean squared error) estimating the importance of viral (vOTU) and bacterial (MAG) communities in explaining variation in nutrient cycling during HTC (higher MSE% values imply increased importance of a given predictor). Random forest mean predictor importance measure was computed for each tree and averaged over the forest (5000 trees). b Significant non-linear correlations between bacterial (black line) and viral (blue line) community activity with nutrient turnover (based on all biogeochemical parameters) during HTC based on metatranscriptomic datasets. c Partial least-squares path model (PLS-PM) comparing the relative importance of different factors explaining nutrient cycling during HTC. PLS-PM describes the relationships between viral and bacterial communities (beta-dissimilarity based on mesophilic and thermophilic MAGs or vOTU), viral and MAG catabolic activity. Strengths of path coefficients are shown as arrow width and numbers beside them, while blue, red and gray colors indicate negative, positive, and non-significant effects, respectively. Path coefficients and coefficients of determination (R²) were calculated after 999 bootstraps and significance levels are indicated when p < 0.05. In (b), gray cloud represents a 95% confidence interval around the predicted values. d Schematic illustration describing the relationships between mesophilic and thermophilic bacteria and their viruses in relation to nutrient cycling during HTC.