Fig. 3.

Boruta random forest feature selection of functions that significantly segregated across treatments based on 1000 permutations for a COG annotation, b KEGG annotation, and c dbCAN annotation. Heatmaps based on the z-scored TPM normalized relative abundances of annotated ORFs from SIP metagenome samples. The description of the functions displayed in the heatmap is detailed in Supplementary Table S1 (COG), Supplementary Table S2 (KEGG), and Supplementary Table S3 (dbCAN). No EPS: incubation without WH15EPS. Unlab.: EPS-incubation containing ¹²C-WH15EPS. Heavy: “heavy fraction” of incubations containing ¹³C-WH15EPS