Fig. 4.

Receiver-operator and precision-recall curves for comparison methods and the multiple instance learning (MIL) model with and without clinical metadata. Curves are displayed for both clinical datasets using five comparison techniques (top panels) for which the average curves from five testing trials are displayed. The MIL microbiome and microbiome-plus-metadata panels show average curves from five testing trials with 95% confidence intervals in parentheses. AUC, area under curve; SVM, support vector machine; LR, logistic regression