Fig. 3

CRC prediction performance is comparable and taxonomically similar across datasets. a Cross-prediction matrix demonstrating random forest AUC based on genus-level relative abundances. LODO (leave-one-dataset-out) refers to the AUC generated by training a model using all but the dataset of the associated column and testing it using the dataset of that column. The datasets contain samples from the following countries: Feng – Austria; Gupta – India; Thomas_a – Italy; Thomas_b – Italy; Vogtmann – USA; Wirbel – Germany; Yachida – Japan; Yu – Hong Kong, China; Zeller – France. b The importance of each genus for the cross-validation performance in each dataset using gini values. Only genera in the top five in at least one dataset are shown