Fig. 3

Comparison of inferences based on gene composition estimated with metagenome prediction tools and metagenome sequencing in each of the 7 datasets. a In this approach, P values of the Wilcoxon test evaluating the null hypothesis for each dataset (see the “Methods” section and Table S1) were calculated for metagenome sequencing and metagenome prediction tools. The P values for genes in common between the two methods were compared using Spearman’s correlation, and the resulting rho was considered as an estimate for the correlation of inference. b Examples showing the correlations between the P values from metagenome prediction tools and metagenome sequencing in Human_KW, chicken, and Soil_LWM datasets. For example, in the Human_KW dataset, genes higher in urban subjects are in the upper-right hand quadrant, and genes lower in urban are in the lower-left hand quadrant. c The correlation of inference between metagenome sequencing and PICRUSt, PICRUSt2, and Tax4Fun for all seven datasets