Fig. 2

Identification of nutritional-ecotype by NEMoE. a Boxplot comparing NEMoE and single-level NEMoE in estimating shared latent classes. The ARI (x-axis) is calculated by comparing the estimated latent class and the true latent class from the data-generating model. In all settings, NEMoE using multiple-level information performs better. b PCA plot of scaled nutrient intake for subjects colored by the two nutrition classes as estimated by NEMoE. Estimated coefficients of the gating network showed high coefficients for sugar, protein:carbohydrate, and moisture. We denote the two nutrition classes as prot-CARB and PROT-carb with low protein-high carbohydrate intake and vice versa. c Scatter plot of genera Fusicatenibacter and Anaerostipes. Left panel shows that Parkinson’s disease and healthy controls in the prot-CARB subcohort roughly separate but there is no such separation in the PROT-carb right panel. d Scatter plot of genera Erysipelotrichaceae UCG-003 and [Ruminococcus] torques group showed a different relationship between Parkinson’s Disease and Healthy Controls in two nutritional-ecotypes