TABLE 2.

Performance of the indicated 2-gene SVM classifiers for COVID-19 diagnosis in the New York RNA-seq cohort^a

2-gene combination	External dataset (n = 553) AUC Trained on UCSF 70% training set	PPV	NPV	Sens	Spec All	Spec No Virus	Spec Other Virus
IFI6, GRINA	0.883	0.725	0.908	0.795	0.871	0.922	0.671
IFI6, C15orf48	0.861	0.699	0.897	0.771	0.858	0.912	0.646
IFI6, GBP5	0.910	0.742	0.924	0.831	0.876	0.916	0.722
IFI44L, GBP5	0.919	0.749	0.929	0.843	0.879	0.919	0.722
IFI44L, PTAFR	0.894	0.659	0.942	0.886	0.804	0.877	0.519
IFI44L, FCGR1A	0.896	0.701	0.929	0.849	0.845	0.883	0.696
HERC6, TNIP3	0.852	0.687	0.915	0.819	0.840	0.899	0.608
HERC6, GBP5	0.866	0.677	0.900	0.783	0.840	0.883	0.671
HERC6, C0A3	0.797	0.563	0.898	0.807	0.731	0.795	0.481

^aThe positive predictive value (PPV), negative predictive value (NPV), sensitivity (Sens), and specificity (Spec) values shown are calculated on Youden’s index when the model is evaluated on the testing set.