Table 3.

This table summarizes the general relationships between hepatic fat fraction (HFF) and the features selected in the random forests for the prediction of HFF. Plots are used to understand the direction of the relationships between the predictors and the outcome, as well as the inter-relationships between the predictors, since random forests do not provide regression coefficients; the partial and interaction plots showing these relationships in detail are in the supplemental figures. Since random forests allow for complex interactions between the predictors, this table notes when there is no evidence of strong 2-way interactions and notes the interacting features when there is evidence of interactions. In order to fully understand the nature of these interactions, please see the supplemental figures.

Group	Selected Features	Adjusted association with HFF	Interactions?
Taxa	g_Varibaculum	Negative	No
	o_RF32	Positive	No
	g_Paraprevotella	Positive	No
	g_Sutterella	Positive	Bilophila
	g_Bilophila	Positive	Bacteroides, Oscillospira,
	g_Oscillospira	Negative	Bilophila
	g_Bacteroides	U-shaped: High HFF with low and high abundance, low HFF with moderate	Bilophila
Diet	% Carbohydrates	Negative	% MUFA
	% Total fat	Positive	% MUFA
	% MUFA	Negative	Carbohydrates,
Comorbidities & Demographics	BMI z-score	Positive	Delivery mode
	Delivery mode	Higher HFF with vaginal birth	BMI
Combined Features	g_Bilophila	Positive	Paraprevotella
	g_Paraprevotella	Positive	Bilophila, % MUFA
	% MUFA	Negative	Paraprevotella
	BMI z-score	Positive