Fig. 3: Gradient boosting model for the prediction of RBP-mediated human cell entry.
From: Receptor-binding proteins from animal viruses are broadly compatible with human cell entry factors
a, AUC plot showing the true positive rate (TPR, that is, fraction of infections correctly predicted by the model, also termed sensitivity or recall) versus the false positive rate (FPR, that is, fraction of real negative infections erroneously predicted as positives by the model, or 1 − specificity). The white dot indicates the values obtained when a model score equal to 0.5 was used as threshold value for making predictions. b, Confusion matrix. Counts of model predictions for the presence/absence or infection versus data labels are shown. c, SHAP summary plot showing the impact of the top 20 features on the model predictions. Each dot represents a data point, coloured based on its feature value (low/high for continuous features, no/yes for binary features). The numbers on the left represent the sum of the absolute SHAP values for each feature, indicating its overall importance in the predictions. SHAP plots for individual viruses are provided in Supplementary Fig. 20. a.a., number of amino acids.
