Fig. 4From: Classification of protein–protein association rates based on biophysical informaticsThe overall performance of neural-network-based classification. We plotted the true positive, true negative, false positive and false negative (a), as well as the specificity, sensitivity, precision and overall accuracy (b) as a function of classification threshold. We also compared the true positive rate with the false positive rate from the classification results under different threshold values. The results correspond to a receiver operating characteristic (ROC) curve, as shown in (c). Finally, we found that the classification accuracy has a positive correlation with the confidence score offered by the neural network model (d)Back to article page