Fig. 4. Performance of RAIN machine-learning models.

a Performance metrics of the three algorithms using the test dataset with Accuracy = (TP + TN)/(TP + FP + TN + FN), Recall = TP/(TP + FN), and Precision = TP/(TP + FP). b–d Receiver-operating characteristic (ROC) curves and corresponding area under the curve (AUC) statistics for each bNAb antigenic site with the test dataset. Each row represents one algorithm, b AD, c DT, and d RF. e Most important features with their scores for each bNAb classified by binding antigenic site using the Random Forest classifier.