Skip to main content

Table 3 Comparison of positive predictive values (PPV) from PolyBayes and ML predictor

From: Application of machine learning in SNP discovery

 

PolyBayes

Probability (P)

TP

FP

PPV

P ≤ 0.60

20

1756

1.1

0.60 < P ≤ 0.70

38

1529

2.4

0.70 < P ≤ 0.80

31

1683

1.8

0.80 < P ≤ 0.90

45

2015

2.2

0.90 < P ≤ 0.95

50

1613

3.0

0.95 < P ≤ 0.97

53

1055

4.8

0.97 < P ≤ 0.99

148

2069

6.7

P = 1.00

1050

5235

16.7

Overall

1435

16955

7.8

 

ML Predictor

 

TP

FP

PPV

Overall

1153

207

84.8

  1. TP: True Positive, FP: False Positive,
  2. Positive predictive value (PPV) = TP/(TP + FP).
  3. The number of true positives in the dataset can be increased by using stringent PolyBayes posterior probability cut-off values. However, even when the posterior probability value is set to the maximum of 1.0 the positive predictive value with PolyBayes is less than 20%. Application of machine learning showed a 5–10 fold increase in the PPV at different PolyBayes posterior probability values.