Journal: iScience
Article Title: Profiling biological effects of microbiome metabolites via machine learning
doi: 10.1016/j.isci.2026.115282
Figure Lengend Snippet: Machine learning model validation using external datasets and prospective in vitro experiments (A) Odds ratios of predicted positives in external validation sets relative to the microbiome background. The HIA model (GUTSY, n = 50) and BBB model (NIAGADS, n = 151) showed significant enrichment (ORs = 6.0 and 2.3, respectively; Fisher’s exact test, p = 2 × 10 −4 and p = 5 × 10 −4 , respectively). (B–D) Prospective in vitro validation of drug-induced liver injury prediction in HepG2 cells. (C) Predicted liver-toxic metabolites exhibited dose-dependent, significant cytotoxic effects. (D) Predicted liver-safe metabolites showed no significant cytotoxic effect ( p > 0.05) at any of the tested concentrations up to 1 mM (unpaired t test; N = 3, n = 3). (E) Confusion matrix for the prospective in vitro validation. (F) Evaluation of performance of machine learning models predicting IL-8 secretion stimulation. The RF model outperformed each of the other models (∗ p ≤ 0.05, ∗∗ p ≤ 0.01, ∗∗∗ p ≤ 0.001, ∗∗∗∗ p ≤ 0.0001). (G) Chemical structures of top predicted candidates, spermine and spermidine. (H) Results from the IL-8 secretion assay using human PBMCs, evaluating the effect of seven microbiome metabolites at 100 μM ( N = 1, n = 3). Metabolites stimulating IL-8 are shown in pink, and metabolites with no IL-8 secretion are shown in blue (∗ p ≤ 0.05, ∗∗ p ≤ 0.01, ∗∗∗ p ≤ 0.001, ∗∗∗∗ p ≤ 0.0001). (I) IL-8 stimulation by spermine and spermidine at different concentrations ( N = 2, n ≥ 2). Data are represented as the mean ± standard deviation.
Article Snippet: Primary human peripheral blood mononuclear cell (PBMC) , ATCC , PCS-800-011, Lot 8032322.
Techniques: Biomarker Discovery, In Vitro, Standard Deviation
ATCC is a verified supplier 
ATCC manufactures this product 
