Figure 3

Performances of the top five ML models in the three encoding methods. We illustrated the performances of the top five ML models in three encoding methods by ROC curves with the AUC values. (A) ML analysis using sequence-based features. (B) ML analysis using atom-based features. (C) ML analysis using fingerprint-based features. We found that the XGB model and fingerprint encoding method are the most suitable approaches for classifying neutralizing and nonneutralizing antibody classes.