FIG 3.

Pooling random forest models and feature cross talk. (a) Random forest models in predicting adenoma and carcinoma using stool samples with 10-fold cross-validation. (b and c) Pooling random forest models in predicting adenoma and carcinoma using tissue samples. (d to f) Genus-, species-, and family-level features’ ranks were obtained according to the random forest model’s MDIA and plotted in a heat map, with dark steel blue for the highest rank, white for the lowest rank, and gray for those that were unavailable. All the strategy codes were carried on as described above. All specificity and sensitivity thresholds were decided by the Youden index. NA, not available.