Fig. 4

Identification of fungal OTU-based signatures of early-stage LUAD. The Boruta algorithm was first used to select 19 OTUs as the final features, and five different supervised ML algorithms were used for identifying patients with LUAD based on OTU features in the discovery cohort. A, B Accuracy performance (A) and receiver operating characteristic curves (B) of LR, NBs, KNN, SVM, and RF algorithms. RF achieved the highest accuracy of 86.17% and the maximum AUC of 0.9350. C Importance of the 19 OTU features was ranked using a classification error loss function