Table 2.

Phenotyping accuracy measures at thresholds selected to achieve TPR = 0.8 or PPV = 0.8

	TPR = 0.8				PPV = 0.8
	Threshold	FPR	PPV	NPV	Threshold	TPR	FPR	NPV
Ideal learning	0.63	0.007 (0.001)	0.86 (0.02)	0.99 (0.002)	0.50	0.90 (0.02)	0.01 (0.002)	0.99 (0.001)
Naive logit	0.15	0.009 (0.002)	0.83 (0.03)	0.99 (0.002)	0.14	0.84 (0.03)	0.01 (0.002)	0.99 (0.002)
EN algorithm	0.42	0.009 (0.003)	0.84 (0.05)	0.99 (0.003)	0.38	0.84 (0.04)	0.01 (0.004)	0.99 (0.002)
ML method	0.63	0.007 (0.002)	0.86 (0.03)	0.99 (0.002)	0.50	0.89 (0.03)	0.01 (0.002)	0.99 (0.002)

Values are mean (empirical standard error) over 1000 iterations.

EN: Elkan and Noto; FPR: false positive rate; ML: maximum likelihood; NPV: negative predictive value; PPV: positive predictive value; TPR: true positive rate.