Fig. 5

Association between the gut microbiota and serum inflammatory cytokines and the random forest classifier for CADASIL. A Heat map of the Spearman’s correlation between the abundance of species and six serum inflammatory cytokines. Red indicates the positive correlation, while blue indicates the negative correlation. *p-value < 0.05; **p-value < 0.01. Arrows point out the correlation of three species that were more abundant in the case group and corresponding inflammatory cytokines. B–C The Spearman’s correlation between the abundance of Fusobacterium varium and serum level of IL-1β (B) and IL-6 (C). Samples in orange and blue represent they are from the case group and the control group, respectively. Spearman’s rank correlation coefficient and p-values were shown. D The random forest model was trained using the abundance of fecal metagenomics species, the level of serum cytokines, and targeted neurotransmitters in our cohort. All variables were first ranked based on their variable importance and then added sequentially into the model. The error curves were plotted for the five trials of tenfold cross-validation in random forest classification as the number of variables increased. The black curve indicates the average cross-validation error of the five trials (in gray). The minimum error in the averaged curve plus the standard deviation at that point was used as the cutoff for feature selection. The model containing the smallest number of variables with an error below that cutoff was chosen as the optimal classifier. The red line marks the number of variables in the optimized model (n = 10). E The mean decrease accuracy (MDA) score of potential ten markers in the model. The colors represent enrichment or increase in CADASIL (orange shades). F Receiver operating characteristic (ROC) curve and the performance of the random forest model using the metagenomic species (green) and further a combination with serum cytokines and targeted neurotransmitters (blue)