Fig. 2

Top ranking features (species) during random forest classification training (128 trees) when the overlap dataset was used. Features were identified in terms of random forest importance scores, indicating their contribution to classification performance for a given class. The relative abundances (RPK) for each top ranking feature across all samples were binned (x-axis). The frequency of each feature across samples falling into these bins is shown (y-axis). Bars shaded red indicate the highest ranking feature for a given class. High ranking features with large frequencies at bin 0 suggest that those features are rare, but if present, highly influence the classifier to classify a sample in that feature’s corresponding class. a Surface. b Region