Fig. 4. Random Forest classification identifies metabolites within amino acid, nucleotide, and lipid superpathways that discriminate infections with BV-associated bacteria in 3-D cervical cells.

a Metabolites most predictive of infection with L. parvula, F. gonidiaformans, F. nucleatum, P. lacrimalis, and P. uenonis include acetylated amino acid derivatives (N-acetylasparagine and N1,N1-diacetylspermine), histidine catabolites (imidazole propionate, formiminoglutamate, and trans-urocanate), adenosine derivatives (3’-AMP and adenosine), glycerophospholipids (1-palmitoyl-2-oleyol-GPE and 1-stearoyl-2-oleoyl-GPE), oncometabolite (2-hydroxyglutarate), polyamines (MTA and agmatine), and short-chain fatty acids (2-hydroxybutyrate, butyrate/isobutyrate, and alpha-hydroxyisocaproate). Only significantly different metabolites (infection vs. mock-infected controls) are colored in the heat map. b Random Forest confusion matrix. Individual cells are labeled according to how the algorithm predicted each treatment. Uninfected mock controls and infection with F. gonidiaformans, F. nucleatum, and P. uenonis were perfectly predicted. The number of replicates (n) included for each treatment are indicated in parentheses on the vertical axis labels.