Performance measures

Morten Nielsen,CBS, Department of Systems Biology,

DTU

Performance measures

Meas Pred0.4050 0.83440.9373 1.00000.8161 0.63880.6752 0.98410.0253 0.00000.3196 0.53880.6764 0.62470.1872 0.19210.4220 0.65460.6545 0.65460.7917 0.13420.4405 0.35510.1548 0.00000.2740 0.19930.4399 0.64610.1725 0.39160.0539 0.00000.3795 0.56230.2242 0.19680.3108 0.21140.2260 0.03360.2780 0.56470.0198 0.12240.5890 0.55380.5120 0.43490.7266 1.00000.1136 0.00000.0456 0.21280.0069 0.41000.4502 0.3848

Pearsons’ correlation coefficient

Mean Squared Error

Quantitative data

• Accuracy of prediction method

Sort

Matthews correlation - Threshold of 0.5

False negative

True positive

True negative

False positive

Matthews correlation - Threshold of 0.5

5 9

8 5

Performance measure – Roc curve

False negative

True positive

True negative

False positive

4 10 1 12 0.080.29

Performance measure – Roc curve

False negative

True positive

True negative

False positive

4 10 1 12 0.080.298 6 3 10 0.230.57

Threshold TP FN

TP/(TP+FN) FP TN FP/(FP+TN)

>0.8  4  10  0.29  1  12  0.08

>0.6 8  6  0.57  3  10  0.23 

>0.4 11  3  0.79  6  7  0.46 

>0.2 13 1  0.93  9  4  0.69 

>0 14  0  1  13  0  1 

AUC = 0.5AUC = 1.0

ROC curves

AUC (area under the ROC curve)

Summary

• MCC and PCC• Random = 0.0• Perfect = 1.0 (or -1)

• ROC (AUC)• Random = 0.5• Perfect = 1 (or 0)