Extended Data Fig. 7: RandomForests variable selection (VSURF) on bacterial phyla relative abundance, bacterial and archaeal amoA and 16 S genes abundance, alpha-diversity, soil chemistry and microbial biomass-C to select the variables explaining best the difference in detrended CO2 production rates.
From: Carbon and nitrogen cycling in Yedoma permafrost controlled by microbial functional limitations
Importance is non-normalized % Increase in Mean Squared Error of a tree when the variable is randomly permuted in out-of-bag (OOB) samples (that is a higher value indicates a higher importance), variables are ranked by decreasing importance. OOBinterpretation is the out-of-bag error of the nested forests (that is grown using this variable as well as all variables with greater importance), the VSURF algorithm selects variables leading to the lowest OOBinterpretation score. Variables in grey were considered uninformative at the thresholding phase, variables in bold were selected at the interpretation phase and are termed ‘Community + function’ in Table 1. Variables in bold and italics are included in the multiple linear regression models presented in Table 1.
