Table 1.
Overview of variables used to construct models.
OTU = Operational taxonomic unit.
Fig 1.
Mean absolute error (MAE) from 100 iterations of each respective model against the testing dataset.
Data are reported by biological marker (column), while color compares models with (gold) and without (gray) environmental predictors. Error bars are the standard error of MAE values across all 100 iterations for each respective model.
Fig 2.
Model predictions for the training set (A-C) and testing set (D-F) for the top performing model for each biological marker as determined by the lowest MAE.
For each biological marker, top models included the 16S phylum + environmental data (A, D, line color—red) 16-ITS order (B, E, line color—blue), and ITS order + environmental data (C, F, line color—magenta). Predictability of each model is greater for the training set (A-C) compared to the testing set (D-F). Soild (training set) and dashed (testing set) lines show the best fit linear relationship and shading indicted the 95% confidence interval between actual PMI, in ADH, and predicted ADH within each respective dataset.
Table 2.
Number of microbial taxonomic features provided as input for random forest regression investigated in this study.
16S-ITS microbial features are the sum of 16S and ITS features.
Fig 3.
Mean absolute error (MAE) varies as a result of biological marker (16S, 16S-ITS, or ITS) used for model construction.
Average MAE is the result of 100 iterations of the 24 respective models against the testing set. Reported p-values are the result of post-hoc t-tests adjusted for multiple comparisons with the Holm method.
Fig 4.
Mean absolute error (MAE) did not vary as a result of taxonomic level (color) used for model construction for any of the biological markers assessed (column).
Mean MAE is the result of 100 iterations of the 24 respective models against the testing set. Order level models generally had the lowest MAE, compared to phylum, class, and OTU models. ANOVA p-values are the result of linear models comparing mean MAE to taxonomic level.
Fig 5.
Top 25 model features determined by variance of responses in Ranger.
For each biological marker, top models included 16S phylum + environmental data (A), ITS order + environmental data (B), and 16S-ITS order (C). Bar color denotes whether the feature is a 16S taxon (green), ITS taxon (orange), or environmental feature (purple).
Fig 6.
Relative abundance of the 5 most important bacterial phyla in the top 16S random forest model (16S phylum + environmental data).
Relative abundance of the phyla Firmicutes, Acidobacteria, Epsilonbacteraeota, Proteobacteria, and Nitrospirae change over time, here accumulated degree hours (ADH), within decomposition-impacted soils. Abundances for each of the 19 individuals (named “TOX###”) are delineated by color.
Fig 7.
Relative abundance of the fungal order Pleosporales over time, here accumulated degree hours (ADH).
Abundances for each of the 19 individuals (named “TOX###”) are delineated by color.