Table 2.

Machine learning performance of various lipid panels.

	10-Lipid panel training set	10-Lipid panel test set	17-Lipid panel training set	17-Lipid panel test set	AutoML 17-lipid panel test set
	OC: n = 144, Non-OC: n = 144	OC: n = 64, Non-OC: n = 34	OC: n = 144, Non-OC: n = 144	OC: n = 64, Non-OC: n = 34	OC: n = 64, Non-OC: n = 34
AUC	0.85 ± 0.07	0.82	0.88 ± 0.07	0.83	0.85
Accuracy	0.77 ± 0.09	0.78	0.81 ± 0.09	0.76	0.78
Sensitivity	0.78 ± 0.15	0.78	0.83 ± 0.09	0.75	0.75
Specificity	0.76 ± 0.12	0.76	0.79 ± 0.12	0.76	0.82

Note: Classification performance using a random forest classier and the AutoML model. The top 7 discriminant gangliosides were added to the 10-lipid panel to create the 17-lipid biomarker panel. Training data were balanced with the Synthetic Minority Oversampling Technique (SMOTE).