Table 4 Performance metrics used to evaluate classifiers

Sensitivity (SENS)

TP/(TP + FN)

Measures the ability of a QSAR tool to detect Ames positives compounds correctly.

Specificity (SPEC)

TN/(TN + FN)

Measures the ability for a QSAR tool to detect negatives compounds.

Accuracy (ACC)

(TP + TN)/(TP + TN + FP + FN)

Assesses a QSAR tool’s overall performance by returning the fraction of compounds which were correctly predicted.

Balanced Accuracy (BA)

(SENS + SPEC)/2

Assesses the overall model performance, giving each class equal weight.

Positive Prediction Value (PPV)

(TP)/(TP + FP)

Indicates how frequently positive predictions are correct.

Negative Prediction Value (NPV)

TN/(TN + FN)

Indicates how often negative predictions are correct.

Mathews Correlation Coefficient (MCC)

\( \frac{\left(\left( TP\ast TN\right)-\left( FP\ast FN\right)\right)}{\sqrt{\left( TP+ FP\left)\left( TP+ FN\right)\left( TN+ FP\right)\right( TN+ FN\right)}} \)

Assesses the overall performance of the model. Values can range from −1 to 1, which is in contrast to the other metrics in this table which range form 0 to 1.

Coverage (COV)

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

Evaluates the proportion of compounds for which the model can make a positive or negative prediction.