TABLE 1.

Random forest regression modeling of amplicon data using features collapsed at different taxonomic levels^a

Amplicon	Season(s)	Most accurate level	Range of MAEs	Top five important features	Range of importance
16S rRNA	Spring and summer (“combined”)	ASV	793.33–851.41	Phyllobacteriaceae, Defluvibacter, Corynebacterium, Shinella, Devosia	0.040–0.023
16S rRNA	Spring	ASV	872.02–1,074.76	Gallicola, Cellulosimicrobium, Brachybacterium, Comamonas, Leucobacter	0.075–0.042
16S rRNA	Summer	ASV	723.98–853.38	Phyllobacteriaceae, Sphingopyxis, Alcaligenaceae, Devosia, Pseudaminobacter	0.014–0.010
18S rRNA	Spring and summer (“combined”)	8	941.22–1,128.13	Eurotiomycetes, Sordariomycetes, Metazoa, Saccharomycetes, Tremellomycetes	0.067–0.033
18S rRNA	Spring	5	1,025.53–1,443.86	Mucoromycota, Metazoa, Vannellida, Eumetazoa, Dikarya	0.102–0.047
18S rRNA	Summer	7	820.67–1,083.95	Nematoda, Saccharomycotina, BOLA868, Alveolata, Eumetazoa	0.071–0.037

^aModel accuracy is assessed using mean absolute error (MAE). The range of MAEs resulting from modeling at all taxonomic levels is reported. The top five most important features within each model are arranged from the most to least important, as determined by the random forest regression. Note that some important features were not able to be classified all the way down to the same taxonomic level at which the model was performed (e.g., Metazoa). Underlined features include those commonly important between model types. Note that there are no commonly important features in the 18S rRNA models due differences in the most accurate levels, whereas in the 16S rRNA models, all of the most accurate models were at the ASV level. MAEs for all levels are reported in Table S2.