Fig. 2

Receiver Operating Characteristic (ROC) curves measuring the performance of the Machine Learning classification model at all classification thresholds of 1 average scores and 2 per-class. Model optimization curves are the true positive rates over the range of false positive rates. 3 Model accuracy and 4 overall and baseline accuracy for the Extra-Tree classification of the merged dataset samples for adults with Prader–Willi syndrome (PWS) and adult cohorts (A) residing in Canada, (B) with irritable bowel syndrome residing in Canada, (C) residing in Florida, USA, and (D) older women residing in Florida. 5 Heatmap visualization depicting the relative abundances from fecal samples of the important taxa used for the classification of adults with Prader–Willi syndrome versus non-affected adult groups